fix overlap bug

.gitignore

@@ -11,4 +11,5 @@ test/
 *.log
 /data_generation/data/*
 /data_generation/output/*
-test/
+test/
+temp*

Readme.md

@@ -0,0 +1,192 @@
+# Next Best View for Reconstruction
+
+## 1. Setup Environment
+### 1.1 Install Main Project
+```bash
+mkdir nbv_rec
+cd nbv_rec
+git clone https://git.hofee.top/hofee/nbv_reconstruction.git
+```
+### 1.2 Install PytorchBoot
+the environment is based on PytorchBoot, clone and install it from [PytorchBoot](https://git.hofee.top/hofee/PyTorchBoot.git)
+```bash
+git clone https://git.hofee.top/hofee/PyTorchBoot.git
+cd PyTorchBoot
+pip install .
+cd ..
+```
+### 1.3 Install Blender (Optional)
+If you want to render your own dataset as described in [section 2. Render Datasets](#2-render-datasets), you'll need to install Blender version 4.0 from [Blender Release](https://download.blender.org/release/Blender4.0/). Here is an example of installing Blender on Ubuntu:
+```bash
+wget https://download.blender.org/release/Blender4.0/blender-4.0.2-linux-x64.tar.xz
+tar -xvf blender-4.0.2-linux-x64.tar.xz
+```
+If blender is not in your PATH, you can add it by:
+```bash
+export PATH=$PATH:/path/to/blender/blender-4.0.2-linux-x64
+```
+To run the blender script, you need to install the `pyyaml` and `scipy` package into your blender python environment. Run the following command to print the python path of your blender:
+```bash
+./blender -b --python-expr "import sys; print(sys.executable)"
+```
+Then copy the python path `/path/to/blender_python` shown in the output and run the following command to install the packages:
+```bash
+/path/to/blender_python -m pip install pyyaml scipy
+```
+### 1.4 Install Blender Render Script (Optional)
+Clone the script from [nbv_rec_blender_render](https://git.hofee.top/hofee/nbv_rec_blender_render.git) and rename it to `blender`:
+```bash
+git clone https://git.hofee.top/hofee/nbv_rec_blender_render.git
+mv nbv_rec_blender_render blender
+```
+
+### 1.5 Check Dependencies
+Switch to the project root directory and run `pytorch-boot scan` or `ptb scan` to check if all dependencies are installed:
+```bash
+cd nbv_reconstruction
+pytorch-boot scan
+# or
+ptb scan
+```
+If you see project structure information in the output, it means all dependencies are correctly installed. Otherwise, you may need to run `pip install xxx` to install the missing packages.
+
+## 2. Render Datasets (Optional)
+### 2.1 Download Object Mesh Models
+Download the mesh models divided into three parts from: 
+ - [object_meshes_part1.zip](None)
+ - [object_meshes_part2.zip](https://pan.baidu.com/s/1pBPhrFtBwEGp1g4vwsLIxA?pwd=1234)
+ - [object_meshes_part3.zip](https://pan.baidu.com/s/1peE8HqFFL0qNFhM5OC69gA?pwd=1234)
+
+or download the whole dataset from [object_meshes.zip](https://pan.baidu.com/s/1ilWWgzg_l7_pPBv64eSgzA?pwd=1234)
+
+Download the table model from [table.obj](https://pan.baidu.com/s/1sjjiID25Es_kmcdUIjU_Dw?pwd=1234)
+
+### 2.2 Set Render Configurations
+Open file `configs/local/view_generate_config.yaml` and modify the parameters to fit your needs. You are required to at least set the following parameters in `runner-generate`:
+ - `object_dir`: the directory of the downloaded object mesh models
+ - `output_dir`: the directory to save the rendered dataset
+ - `table_model_path`: the path of the downloaded table model
+
+### 2.3 Render Dataset
+
+There are two ways to render the dataset:
+
+#### 2.3.1 Render with Visual Monitoring
+
+If you want to visually monitor the rendering progress and machine resource usage:
+
+1. In the terminal, run:
+   ```
+   ptb ui
+   ```
+2. Open your browser and visit http://localhost:5000
+3. Navigate to `Project Dashboard - Project Structure - Applications - generate_view`
+4. Click the `Run` button to execute the rendering script
+
+#### 2.3.2 Render in Terminal
+
+If you don't need visual monitoring and prefer to run the rendering process directly in the terminal, simply run:
+
+```
+ptb run generate_view
+```
+
+This command will start the rendering process without launching the UI.
+
+## 3. Preprocess
+
+⚠️ The preprocessing code is currently not managed by `PytorchBoot`. To run the preprocessing:
+
+1. Open the `./preprocess/preprocessor.py` file.
+2. Locate the `if __name__ == "__main__":` block at the bottom of the file.
+3. Specify the dataset folder by setting `root = "path/to/your/dataset"`.
+4. Run the preprocessing script directly:
+
+   ```
+   python ./preprocess/preprocessor.py
+   ```
+
+This will preprocess the data in the specified dataset folder.
+
+## 4. Generate Strategy Label
+
+### 4.1 Set Configuration
+
+Open the file `configs/local/strategy_generate_config.yaml` and modify the parameters to fit your needs. You are required to at least set the following parameter:
+
+- `datasets.OmniObject3d.root_dir`: the directory of your dataset
+
+### 4.2 Generate Strategy Label
+
+There are two ways to generate the strategy label:
+
+#### 4.2.1 Generate with Visual Monitoring
+
+If you want to visually monitor the generation progress and machine resource usage:
+
+1. In the terminal, run:
+   ```
+   ptb ui
+   ```
+2. Open your browser and visit http://localhost:5000
+3. Navigate to Project Dashboard - Project Structure - Applications - generate_strategy
+4. Click the `Run` button to execute the generation script
+
+#### 4.2.2 Generate in Terminal
+
+If you don't need visual monitoring and prefer to run the generation process directly in the terminal, simply run:
+
+```
+ptb run generate_strategy
+```
+
+This command will start the strategy label generation process without launching the UI.
+
+## 5. Train
+
+### 5.1 Set Configuration
+
+Open the file `configs/local/train_config.yaml` and modify the parameters to fit your needs. You are required to at least set the following parameters in the `experiment` section:
+
+```yaml
+experiment:
+  name: your_experiment_name
+  root_dir: path/to/your/experiment_dir
+  use_checkpoint: False # if True, the checkpoint will be loaded
+  epoch: 600 # specific epoch to load, -1 stands for last epoch
+  max_epochs: 5000 # maximum epochs to train
+  save_checkpoint_interval: 1 # save checkpoint interval
+  test_first: True # if True, test process will be performed before training at each epoch
+```
+
+Adjust these parameters according to your training requirements.
+
+
+### 5.2 Start Training
+
+There are two ways to start the training process:
+
+#### 5.2.1 Train with Visual Monitoring
+
+If you want to visually monitor the training progress and machine resource usage:
+
+1. In the terminal, run:
+   ```
+   ptb ui
+   ```
+2. Open your browser and visit http://localhost:5000
+3. Navigate to Project Dashboard - Project Structure - Applications - train
+4. Click the `Run` button to start the training process
+
+#### 5.2.2 Train in Terminal
+
+If you don't need visual monitoring and prefer to run the training process directly in the terminal, simply run:
+
+```
+ptb run train
+```
+
+This command will start the training process without launching the UI.
+
+## 6. Evaluation
+...

TODO.md

@@ -1,22 +0,0 @@
-# TODO
-## 预处理数据
-### 1. 生成view阶段
-**input**: 物体mesh
-
-### 2. 生成label阶段
-**input**: 目标物体点云、目标物体点云法线、桌面扫描点、被拍到的桌面扫描点
-
-**可以删掉的数据**: mask、normal
-
-### 3. 训练阶段
-**input**: 完整点云、pose、label
-
-**可以删掉的数据**：depth
-
-### view生成后
-预处理目标物体点云、目标物体点云法线、桌面扫描点、被拍到的桌面扫描点、完整点云
-
-删除depth、mask、normal
-
-### label生成后
-只上传：完整点云、pose、label

app_heuristic.py

@@ -0,0 +1,8 @@
+from PytorchBoot.application import PytorchBootApplication
+from runners.heuristic import Heuristic
+
+@PytorchBootApplication("exp_heuristic")
+class ExpHeuristic:
+    @staticmethod
+    def start():
+        Heuristic("configs/local/heuristic_exp_config.yaml").run()

app_inference.py

@@ -1,5 +1,6 @@
 from PytorchBoot.application import PytorchBootApplication
 from runners.inferencer import Inferencer
+from runners.inference_server import InferencerServer
 
 @PytorchBootApplication("inference")
 class InferenceApp:
@@ -14,3 +15,17 @@ class InferenceApp:
                 Evaluator("path_to_your_eval_config").run()
         '''       
         Inferencer("./configs/local/inference_config.yaml").run()
+
+@PytorchBootApplication("server")
+class InferenceServerApp:
+    @staticmethod
+    def start():
+        ''' 
+            call default or your custom runners here, code will be executed 
+        automatically when type "pytorch-boot run" or "ptb run" in terminal
+        
+            example:
+                Trainer("path_to_your_train_config").run()
+                Evaluator("path_to_your_eval_config").run()
+        '''       
+        InferencerServer("./configs/server/server_inference_server_config.yaml").run()

app_split.py

@@ -5,5 +5,5 @@ from runners.data_spliter import DataSpliter
 class DataSplitApp:
     @staticmethod
     def start():
-        DataSpliter("configs/server/split_dataset_config.yaml").run()
+        DataSpliter("configs/server/server_split_dataset_config.yaml").run()
         

configs/local/heuristic_exp_config.yaml

@@ -0,0 +1,71 @@
+
+runner:
+  general:
+    seed: 0
+    device: cuda
+    cuda_visible_devices: "0,1,2,3,4,5,6,7"
+    
+  experiment:
+    name: exp_hemisphere_circle_trajectory
+    root_dir: "experiments"
+    epoch: -1 # -1 stands for last epoch
+
+  test:
+    dataset_list:
+      - OmniObject3d_test
+
+  blender_script_path: "/media/hofee/data/project/python/nbv_reconstruction/blender/data_renderer.py"
+  output_dir: "/media/hofee/data/results/nbv_rec_inference/hemisphere_random_241202"
+  voxel_size: 0.003
+  min_new_area: 1.0
+  heuristic_method: hemisphere_random
+
+dataset:
+  # OmniObject3d_train:
+  #   root_dir: "C:\\Document\\Datasets\\inference_test1"
+  #   model_dir: "C:\\Document\\Datasets\\scaled_object_meshes"
+  #   source: seq_reconstruction_dataset_preprocessed
+  #   split_file: "C:\\Document\\Datasets\\data_list\\sample.txt"
+  #   type: test
+  #   filter_degree: 75
+  #   ratio: 1
+  #   batch_size: 1
+  #   num_workers: 12
+  #   pts_num: 8192
+  #   load_from_preprocess: True
+
+  OmniObject3d_test:
+    root_dir: "/media/hofee/data/data/new_testset_output"
+    model_dir: "/media/hofee/data/data/scaled_object_meshes"
+    source: seq_reconstruction_dataset_preprocessed
+    # split_file: "C:\\Document\\Datasets\\data_list\\OmniObject3d_test.txt"
+    type: test
+    filter_degree: 75
+    eval_list:
+      - pose_diff
+      - coverage_rate_increase
+    ratio: 0.1
+    batch_size: 1
+    num_workers: 12
+    pts_num: 8192
+    load_from_preprocess: True
+
+heuristic_methods:
+  hemisphere_random:
+    center: [0, 0, 0]
+    radius_fixed: True
+    fixed_radius: 0.6
+    min_radius: 0.4
+    max_radius: 0.8
+
+  hemisphere_circle_trajectory:
+    center: [0, 0, 0]
+    radius_fixed: False
+    fixed_radius: 0.6
+    min_radius: 0.4
+    max_radius: 0.8
+    phi_list: [15, 45, 75]
+    circle_times: 12
+
+
+

configs/local/inference_config.yaml

@@ -1,76 +1,72 @@
 
 runner:
   general:
-    seed: 1
+    seed: 0
     device: cuda
     cuda_visible_devices: "0,1,2,3,4,5,6,7"
     
   experiment:
-    name: w_gf_wo_lf_full
+    name: train_ab_partial
     root_dir: "experiments"
-    epoch: 1 # -1 stands for last epoch
+    epoch: -1 # -1 stands for last epoch
 
   test:
     dataset_list:
-      - OmniObject3d_train
+      - OmniObject3d_test
 
   blender_script_path: "/media/hofee/data/project/python/nbv_reconstruction/blender/data_renderer.py"
-  output_dir: "/media/hofee/data/project/python/nbv_reconstruction/nbv_reconstruction/test/inference_global_full_on_testset"
-  pipeline: nbv_reconstruction_global_pts_pipeline
-  
+  output_dir: "/media/hofee/data/results/nbv_rec_inference/partial_241202"
+  pipeline: nbv_reconstruction_pipeline
+  voxel_size: 0.003
+  min_new_area: 1.0
 dataset:
-  OmniObject3d_train:
-    root_dir: "/media/hofee/repository/nbv_reconstruction_data_512"
+  # OmniObject3d_train:
+  #   root_dir: "C:\\Document\\Datasets\\inference_test1"
+  #   model_dir: "C:\\Document\\Datasets\\scaled_object_meshes"
+  #   source: seq_reconstruction_dataset_preprocessed
+  #   split_file: "C:\\Document\\Datasets\\data_list\\sample.txt"
+  #   type: test
+  #   filter_degree: 75
+  #   ratio: 1
+  #   batch_size: 1
+  #   num_workers: 12
+  #   pts_num: 8192
+  #   load_from_preprocess: True
+
+  OmniObject3d_test:
+    root_dir: "/media/hofee/data/data/new_testset_output"
     model_dir: "/media/hofee/data/data/scaled_object_meshes"
-    source: seq_nbv_reconstruction_dataset
-    split_file: "/media/hofee/data/project/python/nbv_reconstruction/nbv_reconstruction/test/test_set_list.txt"
+    source: seq_reconstruction_dataset_preprocessed
+    # split_file: "C:\\Document\\Datasets\\data_list\\OmniObject3d_test.txt"
     type: test
     filter_degree: 75
-    ratio: 1
+    eval_list:
+      - pose_diff
+      - coverage_rate_increase
+    ratio: 0.1
     batch_size: 1
     num_workers: 12
-    pts_num: 4096
-    load_from_preprocess: False
+    pts_num: 8192
+    load_from_preprocess: True
 
 pipeline:
-  nbv_reconstruction_local_pts_pipeline:
+  nbv_reconstruction_pipeline:
     modules:
       pts_encoder: pointnet_encoder
       seq_encoder: transformer_seq_encoder
       pose_encoder: pose_encoder
       view_finder: gf_view_finder
     eps: 1e-5
-    global_scanned_feat: False
-
-  nbv_reconstruction_global_pts_pipeline:
-    modules:
-      pts_encoder: pointnet_encoder
-      pose_seq_encoder: transformer_pose_seq_encoder
-      pose_encoder: pose_encoder
-      view_finder: gf_view_finder
-    eps: 1e-5
     global_scanned_feat: True
 
-
-    
 module:
-
   pointnet_encoder:
     in_dim: 3
     out_dim: 1024
     global_feat: True
     feature_transform: False
-
   transformer_seq_encoder:
-    pts_embed_dim: 1024
-    pose_embed_dim: 256
-    num_heads: 4
-    ffn_dim: 256
-    num_layers: 3
-    output_dim: 2048
-
-  transformer_pose_seq_encoder:
-    pose_embed_dim: 256
+    embed_dim: 320
     num_heads: 4
     ffn_dim: 256
     num_layers: 3
@@ -86,7 +82,8 @@ module:
     sample_mode: ode
     sampling_steps: 500
     sde_mode: ve
-
   pose_encoder:
     pose_dim: 9
     out_dim: 256
+  pts_num_encoder:
+    out_dim: 64

configs/local/strategy_generate_config.yaml

@@ -11,15 +11,10 @@ runner:
     root_dir: "experiments"
       
   generate:
-    voxel_threshold: 0.01
-    soft_overlap_threshold: 0.3
-    hard_overlap_threshold: 0.6
-    filter_degree: 75
-    to_specified_dir: True # if True, output_dir is used, otherwise, root_dir is used
-    save_points: True
-    load_points: True
-    save_best_combined_points: False
-    save_mesh: True
+    voxel_threshold: 0.003
+    overlap_area_threshold: 30
+    compute_with_normal: False
+    scan_points_threshold: 10
     overwrite: False
     seq_num: 10
     dataset_list:
@@ -27,11 +22,6 @@ runner:
   
 datasets:
     OmniObject3d:
-      #"/media/hofee/data/data/temp_output"
-      root_dir: "/media/hofee/repository/new_full_box_data" 
-      model_dir: "/media/hofee/data/data/scaled_object_meshes"
-      from: 0
-      to: -1 # -1 means end
-      #output_dir: "/media/hofee/data/data/label_output"
-
-    
+      root_dir: /data/hofee/nbv_rec_part2_preprocessed
+      from: 155
+      to: 165 # ..-1 means end

configs/local/train_config.yaml

@@ -84,7 +84,7 @@ module:
   gf_view_finder:
     t_feat_dim: 128
     pose_feat_dim: 256
-    main_feat_dim: 2048
+    main_feat_dim: 3072
     regression_head: Rx_Ry_and_T
     pose_mode: rot_matrix
     per_point_feature: False

configs/local/view_generate_config.yaml

@@ -8,30 +8,30 @@ runner:
     root_dir: experiments
   generate:
     port: 5002
-    from: 2200
-    to: 2300 # -1 means all
-    object_dir: /media/hofee/data/data/scaled_object_meshes
-    table_model_path: /media/hofee/data/data/others/table.obj
-    output_dir: /media/hofee/repository/new_data_with_normal
+    from: 1
+    to: 50 # -1 means all
+    object_dir: C:\\Document\\Datasets\\scaled_object_meshes
+    table_model_path: C:\\Document\\Datasets\\table.obj
+    output_dir: C:\\Document\\Datasets\\debug_generate_view
     binocular_vision: true
     plane_size: 10
     max_views: 512
     min_views: 128
-    random_view_ratio: 0.2
+    random_view_ratio: 0.02
     min_cam_table_included_degree: 20
     max_diag: 0.7
-    min_diag: 0.1
+    min_diag: 0.01
     random_config:
       display_table:
         min_height: 0.05
         max_height: 0.15
-        min_radius: 0.3
-        max_radius: 0.5
+        min_radius: 0.2
+        max_radius: 0.3
       display_object:
         min_x: 0
-        max_x: 0.03
+        max_x: 0.05
         min_y: 0
-        max_y: 0.03
+        max_y: 0.05
         min_z: 0.01
         max_z: 0.01
         random_rotation_ratio: 0.3
@@ -43,10 +43,10 @@ runner:
         near_plane: 0.01
         far_plane: 5
         fov_vertical: 25
-        resolution: [1280,800]
-        eye_distance: 0.15
+        resolution: [640,400]
+        eye_distance: 0.10
         eye_angle: 25
       Light:
         location: [0,0,3.5]
         orientation: [0,0,0]
-        power: 150
+        power: 150

configs/server/server_inference_server_config.yaml

@@ -0,0 +1,53 @@
+
+runner:
+  general:
+    seed: 0
+    device: cuda
+    cuda_visible_devices: "0,1,2,3,4,5,6,7"
+    
+  experiment:
+    name: train_ab_global_only
+    root_dir: "experiments"
+    epoch: -1 # -1 stands for last epoch
+
+  pipeline: nbv_reconstruction_pipeline
+  voxel_size: 0.003
+
+pipeline:
+  nbv_reconstruction_pipeline:
+    modules:
+      pts_encoder: pointnet_encoder
+      seq_encoder: transformer_seq_encoder
+      pose_encoder: pose_encoder
+      view_finder: gf_view_finder
+    eps: 1e-5
+    global_scanned_feat: True
+
+module:
+  pointnet_encoder:
+    in_dim: 3
+    out_dim: 1024
+    global_feat: True
+    feature_transform: False
+  transformer_seq_encoder:
+    embed_dim: 256
+    num_heads: 4
+    ffn_dim: 256
+    num_layers: 3
+    output_dim: 1024
+
+  gf_view_finder:
+    t_feat_dim: 128
+    pose_feat_dim: 256
+    main_feat_dim: 2048
+    regression_head: Rx_Ry_and_T
+    pose_mode: rot_matrix
+    per_point_feature: False
+    sample_mode: ode
+    sampling_steps: 500
+    sde_mode: ve
+  pose_encoder:
+    pose_dim: 9
+    out_dim: 256
+  pts_num_encoder:
+    out_dim: 64

configs/server/server_split_dataset_config.yaml

@@ -10,13 +10,13 @@ runner:
     root_dir: "experiments"
   
   split: # 
-    root_dir: "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy"
+    root_dir: "/data/hofee/data/packed_preprocessed_data"
     type: "unseen_instance" # "unseen_category"
     datasets:
       OmniObject3d_train: 
-        path: "../data/sample_for_training_preprocessed/OmniObject3d_train.txt"
+        path: "/data/hofee/data/OmniObject3d_train.txt"
         ratio: 0.9
 
       OmniObject3d_test: 
-        path: "../data/sample_for_training_preprocessed/OmniObject3d_test.txt"
+        path: "/data/hofee/data/OmniObject3d_test.txt"
         ratio: 0.1

configs/server/server_strategy_generate_config.yaml

@@ -1,32 +0,0 @@
-
-runner:
-  general:
-    seed: 0
-    device: cpu
-    cuda_visible_devices: "0,1,2,3,4,5,6,7"
-
-    
-  experiment:
-    name: debug
-    root_dir: "experiments"
-      
-  generate:
-    voxel_threshold: 0.01
-    overlap_threshold: 0.5
-    filter_degree: 75
-    to_specified_dir: False # if True, output_dir is used, otherwise, root_dir is used
-    save_points: True
-    save_best_combined_points: True
-    save_mesh: True
-    overwrite: False
-    dataset_list:
-      - OmniObject3d
-  
-datasets:
-    OmniObject3d:
-      #"/media/hofee/data/data/temp_output"
-      root_dir: "../data/sample_for_training/scenes" 
-      model_dir: "../data/scaled_object_meshes"
-      #output_dir: "/media/hofee/data/data/label_output"
-
-    

configs/server/server_train_config.yaml

@@ -3,13 +3,13 @@ runner:
   general:
     seed: 0
     device: cuda
-    cuda_visible_devices: "1"
+    cuda_visible_devices: "0"
     parallel: False
     
   experiment:
-    name: full_w_global_feat_wo_local_pts_feat
+    name: train_ab_global_only
     root_dir: "experiments"
-    use_checkpoint: False
+    use_checkpoint: True
     epoch: -1 # -1 stands for last epoch
     max_epochs: 5000
     save_checkpoint_interval: 1
@@ -28,57 +28,57 @@ runner:
       - OmniObject3d_test
       - OmniObject3d_val
 
-  pipeline: nbv_reconstruction_global_pts_pipeline
+  pipeline: nbv_reconstruction_pipeline
   
 dataset:
   OmniObject3d_train:
-    root_dir: "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy"
+    root_dir: "/data/hofee/data/new_full_data"
     model_dir: "../data/scaled_object_meshes"
     source: nbv_reconstruction_dataset
-    split_file: "/home/data/hofee/project/nbv_rec/data/OmniObject3d_train.txt"
+    split_file: "/data/hofee/data/new_full_data_list/OmniObject3d_train.txt"
     type: train
     cache: True
     ratio: 1
-    batch_size: 160
-    num_workers: 16
-    pts_num: 4096
+    batch_size: 80
+    num_workers: 128
+    pts_num: 8192
     load_from_preprocess: True
 
   OmniObject3d_test:
-    root_dir: "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy"
+    root_dir: "/data/hofee/data/new_full_data"
     model_dir: "../data/scaled_object_meshes"
     source: nbv_reconstruction_dataset
-    split_file: "/home/data/hofee/project/nbv_rec/data/OmniObject3d_test.txt"
+    split_file: "/data/hofee/data/new_full_data_list/OmniObject3d_test.txt"
     type: test
     cache: True
     filter_degree: 75
     eval_list:
       - pose_diff
-    ratio: 0.05
-    batch_size: 160
+    ratio: 1
+    batch_size: 80
     num_workers: 12
-    pts_num: 4096
+    pts_num: 8192
     load_from_preprocess: True
 
   OmniObject3d_val:
-    root_dir: "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy"
+    root_dir: "/data/hofee/data/new_full_data"
     model_dir: "../data/scaled_object_meshes"
     source: nbv_reconstruction_dataset
-    split_file: "/home/data/hofee/project/nbv_rec/data/OmniObject3d_train.txt"
+    split_file: "/data/hofee/data/new_full_data_list/OmniObject3d_train.txt"
     type: test
     cache: True
     filter_degree: 75
     eval_list:
       - pose_diff
-    ratio: 0.005
-    batch_size: 160
+    ratio: 0.1
+    batch_size: 80
     num_workers: 12
-    pts_num: 4096
+    pts_num: 8192
     load_from_preprocess: True
 
 
 pipeline:
-  nbv_reconstruction_local_pts_pipeline:
+  nbv_reconstruction_pipeline:
     modules:
       pts_encoder: pointnet_encoder
       seq_encoder: transformer_seq_encoder
@@ -87,16 +87,6 @@ pipeline:
     eps: 1e-5
     global_scanned_feat: True
 
-  nbv_reconstruction_global_pts_pipeline:
-    modules:
-      pts_encoder: pointnet_encoder
-      pose_seq_encoder: transformer_pose_seq_encoder
-      pose_encoder: pose_encoder
-      view_finder: gf_view_finder
-    eps: 1e-5
-    global_scanned_feat: True
-
-
     
 module:
 
@@ -107,15 +97,7 @@ module:
     feature_transform: False
 
   transformer_seq_encoder:
-    pts_embed_dim: 1024
-    pose_embed_dim: 256
-    num_heads: 4
-    ffn_dim: 256
-    num_layers: 3
-    output_dim: 2048
-
-  transformer_pose_seq_encoder:
-    pose_embed_dim: 256
+    embed_dim: 256
     num_heads: 4
     ffn_dim: 256
     num_layers: 3
@@ -136,6 +118,9 @@ module:
     pose_dim: 9
     out_dim: 256
 
+  pts_num_encoder:
+    out_dim: 64
+
 loss_function:
   gf_loss:
 

configs/server/server_view_generate_config.yaml

@@ -1,53 +0,0 @@
-runner:
-  general:
-    seed: 0
-    device: cpu
-    cuda_visible_devices: 0,1,2,3,4,5,6,7
-  experiment:
-    name: debug
-    root_dir: experiments
-  generate:
-    object_dir: ../data/scaled_object_meshes
-    table_model_path: ../data/others/table.obj
-    output_dir: ../data/nbv_reconstruction_data_512
-    binocular_vision: true
-    plane_size: 10
-    max_views: 512
-    min_views: 64
-    max_diag: 0.7
-    min_diag: 0.1
-    random_config:
-      display_table:
-        min_height: 0.05
-        max_height: 0.15
-        min_radius: 0.3
-        max_radius: 0.5
-        min_R: 0.05
-        max_R: 0.3
-        min_G: 0.05
-        max_G: 0.3
-        min_B: 0.05
-        max_B: 0.3
-      display_object:
-        min_x: 0
-        max_x: 0.03
-        min_y: 0
-        max_y: 0.03
-        min_z: 0.01
-        max_z: 0.01
-        random_rotation_ratio: 0.3
-      random_objects:
-        num: 4
-        cluster: 0.9
-    light_and_camera_config:
-      Camera:
-        near_plane: 0.01
-        far_plane: 5
-        fov_vertical: 25
-        resolution: [1280,800]
-        eye_distance: 0.15
-        eye_angle: 25
-      Light:
-        location: [0,0,3.5]
-        orientation: [0,0,0]
-        power: 150

core/global_pts_n_num_pipeline.py

@@ -1,100 +0,0 @@
-import torch
-from torch import nn
-import PytorchBoot.namespace as namespace
-import PytorchBoot.stereotype as stereotype
-from PytorchBoot.factory.component_factory import ComponentFactory
-from PytorchBoot.utils import Log
-
-
-@stereotype.pipeline("nbv_reconstruction_global_pts_n_num_pipeline")
-class NBVReconstructionGlobalPointsPipeline(nn.Module):
-    def __init__(self, config):
-        super(NBVReconstructionGlobalPointsPipeline, self).__init__()
-        self.config = config
-        self.module_config = config["modules"]
-        self.pts_encoder = ComponentFactory.create(namespace.Stereotype.MODULE, self.module_config["pts_encoder"])
-        self.pose_encoder = ComponentFactory.create(namespace.Stereotype.MODULE, self.module_config["pose_encoder"])
-        self.pose_n_num_seq_encoder = ComponentFactory.create(namespace.Stereotype.MODULE, self.module_config["pose_n_num_seq_encoder"])
-        self.view_finder = ComponentFactory.create(namespace.Stereotype.MODULE, self.module_config["view_finder"])
-        self.pts_num_encoder = ComponentFactory.create(namespace.Stereotype.MODULE, self.module_config["pts_num_encoder"])
-        
-        self.eps = float(self.config["eps"])
-        self.enable_global_scanned_feat = self.config["global_scanned_feat"]
-        
-    def forward(self, data):
-        mode = data["mode"]
-        
-        if mode == namespace.Mode.TRAIN:
-            return self.forward_train(data)
-        elif mode == namespace.Mode.TEST:
-            return self.forward_test(data)
-        else:
-            Log.error("Unknown mode: {}".format(mode), True)
-    
-    def pertube_data(self, gt_delta_9d):
-        bs = gt_delta_9d.shape[0]
-        random_t = torch.rand(bs, device=gt_delta_9d.device) * (1. - self.eps) + self.eps
-        random_t = random_t.unsqueeze(-1)
-        mu, std = self.view_finder.marginal_prob(gt_delta_9d, random_t)
-        std = std.view(-1, 1)
-        z = torch.randn_like(gt_delta_9d)
-        perturbed_x = mu + z * std
-        target_score = - z * std / (std ** 2)
-        return perturbed_x, random_t, target_score, std
-    
-    def forward_train(self, data):
-        main_feat = self.get_main_feat(data)
-        ''' get std '''
-        best_to_world_pose_9d_batch = data["best_to_world_pose_9d"]
-        perturbed_x, random_t, target_score, std = self.pertube_data(best_to_world_pose_9d_batch)
-        input_data = {
-            "sampled_pose": perturbed_x,
-            "t": random_t,
-            "main_feat": main_feat,
-        }
-        estimated_score = self.view_finder(input_data)
-        output = {
-            "estimated_score": estimated_score,
-            "target_score": target_score,
-            "std": std
-        }
-        return output
-    
-    def forward_test(self,data):
-        main_feat = self.get_main_feat(data)
-        estimated_delta_rot_9d, in_process_sample = self.view_finder.next_best_view(main_feat)
-        result = {
-            "pred_pose_9d": estimated_delta_rot_9d,
-            "in_process_sample": in_process_sample
-        }
-        return result
-    
-    
-    def get_main_feat(self, data):
-        scanned_n_to_world_pose_9d_batch = data['scanned_n_to_world_pose_9d']
-        scanned_target_pts_num_batch = data['scanned_target_points_num']
-
-        device = next(self.parameters()).device
-
-        embedding_list_batch = []
-        
-        for scanned_n_to_world_pose_9d,scanned_target_pts_num in zip(scanned_n_to_world_pose_9d_batch,scanned_target_pts_num_batch):
-            scanned_n_to_world_pose_9d = scanned_n_to_world_pose_9d.to(device)
-            scanned_target_pts_num = scanned_target_pts_num.to(device)
-            pose_feat_seq = self.pose_encoder.encode_pose(scanned_n_to_world_pose_9d)
-            pts_num_feat_seq = self.pts_num_encoder.encode_pts_num(scanned_target_pts_num)
-            embedding_list_batch.append(torch.cat([pose_feat_seq, pts_num_feat_seq], dim=-1))
-
-        main_feat = self.pose_n_num_seq_encoder.encode_sequence(embedding_list_batch)
-        
-        
-        combined_scanned_pts_batch = data['combined_scanned_pts']   
-        global_scanned_feat = self.pts_encoder.encode_points(combined_scanned_pts_batch)
-        main_feat = torch.cat([main_feat, global_scanned_feat], dim=-1)
-        
-        
-        if torch.isnan(main_feat).any():
-            Log.error("nan in main_feat", True)
-        
-        return main_feat
-    

core/nbv_dataset.py

@@ -8,7 +8,7 @@ import torch
 import os
 import sys
 
-sys.path.append(r"/home/data/hofee/project/nbv_rec/nbv_reconstruction")
+sys.path.append(r"/data/hofee/project/nbv_rec/nbv_reconstruction")
 
 from utils.data_load import DataLoadUtil
 from utils.pose import PoseUtil
@@ -31,7 +31,7 @@ class NBVReconstructionDataset(BaseDataset):
         self.load_from_preprocess = config.get("load_from_preprocess", False)
 
         if self.type == namespace.Mode.TEST:
-            self.model_dir = config["model_dir"]
+            #self.model_dir = config["model_dir"]
             self.filter_degree = config["filter_degree"]
         if self.type == namespace.Mode.TRAIN:
             scale_ratio = 1
@@ -66,7 +66,9 @@ class NBVReconstructionDataset(BaseDataset):
                 if max_coverage_rate > scene_max_coverage_rate:
                     scene_max_coverage_rate = max_coverage_rate
                 max_coverage_rate_list.append(max_coverage_rate)
-            mean_coverage_rate = np.mean(max_coverage_rate_list)
+                
+            if max_coverage_rate_list:
+                mean_coverage_rate = np.mean(max_coverage_rate_list)
 
             for seq_idx in range(seq_num):
                 label_path = DataLoadUtil.get_label_path(
@@ -112,6 +114,10 @@ class NBVReconstructionDataset(BaseDataset):
         except Exception as e:
             Log.error(f"Save cache failed: {e}")
 
+    def voxel_downsample_with_mask(self, pts, voxel_size):
+        pass
+
+
     def __getitem__(self, index):
         data_item_info = self.datalist[index]
         scanned_views = data_item_info["scanned_views"]
@@ -122,65 +128,20 @@ class NBVReconstructionDataset(BaseDataset):
             scanned_views_pts,
             scanned_coverages_rate,
             scanned_n_to_world_pose,
-            scanned_target_pts_num,
-        ) = ([], [], [], [])
-        target_pts_num_dict = DataLoadUtil.load_target_pts_num_dict(
-            self.root_dir, scene_name
-        )
+        ) = ([], [], [])
         for view in scanned_views:
             frame_idx = view[0]
             coverage_rate = view[1]
             view_path = DataLoadUtil.get_path(self.root_dir, scene_name, frame_idx)
             cam_info = DataLoadUtil.load_cam_info(view_path, binocular=True)
-            target_pts_num = target_pts_num_dict[frame_idx]
-            n_to_world_pose = cam_info["cam_to_world"]
-            nR_to_world_pose = cam_info["cam_to_world_R"]
-
-            if self.load_from_preprocess:
-                downsampled_target_point_cloud = (
-                    DataLoadUtil.load_from_preprocessed_pts(view_path)
-                )
-            else:
-                cached_data = None
-                if self.cache:
-                    cached_data = self.load_from_cache(scene_name, frame_idx)
-
-                if cached_data is None:
-                    print("load depth")
-                    depth_L, depth_R = DataLoadUtil.load_depth(
-                        view_path,
-                        cam_info["near_plane"],
-                        cam_info["far_plane"],
-                        binocular=True,
-                    )
-                    point_cloud_L = DataLoadUtil.get_point_cloud(
-                        depth_L, cam_info["cam_intrinsic"], n_to_world_pose
-                    )["points_world"]
-                    point_cloud_R = DataLoadUtil.get_point_cloud(
-                        depth_R, cam_info["cam_intrinsic"], nR_to_world_pose
-                    )["points_world"]
-
-                    point_cloud_L = PtsUtil.random_downsample_point_cloud(
-                        point_cloud_L, 65536
-                    )
-                    point_cloud_R = PtsUtil.random_downsample_point_cloud(
-                        point_cloud_R, 65536
-                    )
-                    overlap_points = PtsUtil.get_overlapping_points(
-                        point_cloud_L, point_cloud_R
-                    )
-                    downsampled_target_point_cloud = (
-                        PtsUtil.random_downsample_point_cloud(
-                            overlap_points, self.pts_num
-                        )
-                    )
-                    if self.cache:
-                        self.save_to_cache(
-                            scene_name, frame_idx, downsampled_target_point_cloud
-                        )
-                else:
-                    downsampled_target_point_cloud = cached_data
-
+            
+            n_to_world_pose = cam_info["cam_to_world"]  
+            target_point_cloud = (
+                DataLoadUtil.load_from_preprocessed_pts(view_path)
+            )
+            downsampled_target_point_cloud = PtsUtil.random_downsample_point_cloud(
+                target_point_cloud, self.pts_num
+            )
             scanned_views_pts.append(downsampled_target_point_cloud)
             scanned_coverages_rate.append(coverage_rate)
             n_to_world_6d = PoseUtil.matrix_to_rotation_6d_numpy(
@@ -189,7 +150,7 @@ class NBVReconstructionDataset(BaseDataset):
             n_to_world_trans = n_to_world_pose[:3, 3]
             n_to_world_9d = np.concatenate([n_to_world_6d, n_to_world_trans], axis=0)
             scanned_n_to_world_pose.append(n_to_world_9d)
-            scanned_target_pts_num.append(target_pts_num)
+
 
         nbv_idx, nbv_coverage_rate = nbv[0], nbv[1]
         nbv_path = DataLoadUtil.get_path(self.root_dir, scene_name, nbv_idx)
@@ -205,30 +166,18 @@ class NBVReconstructionDataset(BaseDataset):
         )
 
         combined_scanned_views_pts = np.concatenate(scanned_views_pts, axis=0)
-        voxel_downsampled_combined_scanned_pts_np = (
-            PtsUtil.voxel_downsample_point_cloud(combined_scanned_views_pts, 0.002)
-        )
-        random_downsampled_combined_scanned_pts_np = (
-            PtsUtil.random_downsample_point_cloud(
-                voxel_downsampled_combined_scanned_pts_np, self.pts_num
-            )
-        )
+        voxel_downsampled_combined_scanned_pts_np = PtsUtil.voxel_downsample_point_cloud(combined_scanned_views_pts, 0.002)
+        random_downsampled_combined_scanned_pts_np = PtsUtil.random_downsample_point_cloud(voxel_downsampled_combined_scanned_pts_np, self.pts_num)
+        
         data_item = {
-            "scanned_pts": np.asarray(scanned_views_pts, dtype=np.float32),
-            "combined_scanned_pts": np.asarray(
-                random_downsampled_combined_scanned_pts_np, dtype=np.float32
-            ),
-            "scanned_coverage_rate": scanned_coverages_rate,
-            "scanned_n_to_world_pose_9d": np.asarray(
-                scanned_n_to_world_pose, dtype=np.float32
-            ),
-            "best_coverage_rate": nbv_coverage_rate,
-            "best_to_world_pose_9d": np.asarray(best_to_world_9d, dtype=np.float32),
-            "seq_max_coverage_rate": max_coverage_rate,
-            "scene_name": scene_name,
-            "scanned_target_points_num": np.asarray(
-                scanned_target_pts_num, dtype=np.int32
-            ),
+            "scanned_pts": np.asarray(scanned_views_pts, dtype=np.float32), # Ndarray(S x Nv x 3)
+            "combined_scanned_pts": np.asarray(random_downsampled_combined_scanned_pts_np, dtype=np.float32), # Ndarray(N x 3)
+            "scanned_coverage_rate": scanned_coverages_rate, # List(S): Float, range(0, 1)
+            "scanned_n_to_world_pose_9d": np.asarray(scanned_n_to_world_pose, dtype=np.float32), # Ndarray(S x 9)
+            "best_coverage_rate": nbv_coverage_rate, # Float, range(0, 1)
+            "best_to_world_pose_9d": np.asarray(best_to_world_9d, dtype=np.float32), # Ndarray(9)
+            "seq_max_coverage_rate": max_coverage_rate, # Float, range(0, 1)
+            "scene_name": scene_name, # String
         }
 
         return data_item
@@ -239,33 +188,30 @@ class NBVReconstructionDataset(BaseDataset):
     def get_collate_fn(self):
         def collate_fn(batch):
             collate_data = {}
+            
+            ''' ------ Varialbe Length ------ '''
+            
             collate_data["scanned_pts"] = [
                 torch.tensor(item["scanned_pts"]) for item in batch
             ]
             collate_data["scanned_n_to_world_pose_9d"] = [
                 torch.tensor(item["scanned_n_to_world_pose_9d"]) for item in batch
             ]
-            collate_data["scanned_target_points_num"] = [
-                torch.tensor(item["scanned_target_points_num"]) for item in batch
-            ]
+            
+            ''' ------ Fixed Length ------ '''
+            
             collate_data["best_to_world_pose_9d"] = torch.stack(
                 [torch.tensor(item["best_to_world_pose_9d"]) for item in batch]
             )
             collate_data["combined_scanned_pts"] = torch.stack(
                 [torch.tensor(item["combined_scanned_pts"]) for item in batch]
             )
-            if "first_frame_to_world" in batch[0]:
-                collate_data["first_frame_to_world"] = torch.stack(
-                    [torch.tensor(item["first_frame_to_world"]) for item in batch]
-                )
             for key in batch[0].keys():
                 if key not in [
                     "scanned_pts",
                     "scanned_n_to_world_pose_9d",
                     "best_to_world_pose_9d",
-                    "first_frame_to_world",
                     "combined_scanned_pts",
-                    "scanned_target_points_num",
                 ]:
                     collate_data[key] = [item[key] for item in batch]
             return collate_data
@@ -281,10 +227,9 @@ if __name__ == "__main__":
     torch.manual_seed(seed)
     np.random.seed(seed)
     config = {
-        "root_dir": "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy",
-        "model_dir": "/home/data/hofee/project/nbv_rec/data/scaled_object_meshes",
+        "root_dir": "/data/hofee/data/packed_preprocessed_data",
         "source": "nbv_reconstruction_dataset",
-        "split_file": "/home/data/hofee/project/nbv_rec/data/OmniObject3d_test.txt",
+        "split_file": "/data/hofee/data/OmniObject3d_train.txt",
         "load_from_preprocess": True,
         "ratio": 0.5,
         "batch_size": 2,

core/old_seq_dataset.py

@@ -0,0 +1,154 @@
+import numpy as np
+from PytorchBoot.dataset import BaseDataset
+import PytorchBoot.namespace as namespace
+import PytorchBoot.stereotype as stereotype
+from PytorchBoot.utils.log_util import Log
+import torch
+import os
+import sys
+sys.path.append(r"/home/data/hofee/project/nbv_rec/nbv_reconstruction")
+
+from utils.data_load import DataLoadUtil
+from utils.pose import PoseUtil
+from utils.pts import PtsUtil
+
+@stereotype.dataset("old_seq_nbv_reconstruction_dataset")
+class SeqNBVReconstructionDataset(BaseDataset):
+    def __init__(self, config):
+        super(SeqNBVReconstructionDataset, self).__init__(config)
+        self.type = config["type"]
+        if self.type != namespace.Mode.TEST:
+            Log.error("Dataset <seq_nbv_reconstruction_dataset> Only support test mode", terminate=True)
+        self.config = config
+        self.root_dir = config["root_dir"]
+        self.split_file_path = config["split_file"]
+        self.scene_name_list = self.load_scene_name_list()
+        self.datalist = self.get_datalist()
+        self.pts_num = config["pts_num"]
+        
+        self.model_dir = config["model_dir"]
+        self.filter_degree = config["filter_degree"]
+        self.load_from_preprocess = config.get("load_from_preprocess", False)
+        
+
+    def load_scene_name_list(self):
+        scene_name_list = []
+        with open(self.split_file_path, "r") as f:
+            for line in f:
+                scene_name = line.strip()
+                scene_name_list.append(scene_name)
+        return scene_name_list
+    
+    def get_datalist(self):
+        datalist = []
+        for scene_name in self.scene_name_list:
+            seq_num = DataLoadUtil.get_label_num(self.root_dir, scene_name)
+            scene_max_coverage_rate = 0
+            scene_max_cr_idx = 0
+
+            for seq_idx in range(seq_num):
+                label_path = DataLoadUtil.get_label_path(self.root_dir, scene_name, seq_idx)
+                label_data = DataLoadUtil.load_label(label_path)
+                max_coverage_rate = label_data["max_coverage_rate"]
+                if max_coverage_rate > scene_max_coverage_rate:
+                    scene_max_coverage_rate = max_coverage_rate
+                    scene_max_cr_idx = seq_idx
+        
+            label_path = DataLoadUtil.get_label_path(self.root_dir, scene_name, scene_max_cr_idx)
+            label_data = DataLoadUtil.load_label(label_path)
+            first_frame = label_data["best_sequence"][0]
+            best_seq_len = len(label_data["best_sequence"])
+            datalist.append({
+                    "scene_name": scene_name,
+                    "first_frame": first_frame,
+                    "max_coverage_rate": scene_max_coverage_rate,
+                    "best_seq_len": best_seq_len,
+                    "label_idx": scene_max_cr_idx,
+                })
+        return datalist
+
+    def __getitem__(self, index):
+        data_item_info = self.datalist[index]
+        first_frame_idx = data_item_info["first_frame"][0]
+        first_frame_coverage = data_item_info["first_frame"][1]
+        max_coverage_rate = data_item_info["max_coverage_rate"]
+        scene_name = data_item_info["scene_name"]
+        first_cam_info = DataLoadUtil.load_cam_info(DataLoadUtil.get_path(self.root_dir, scene_name, first_frame_idx), binocular=True)
+        first_view_path = DataLoadUtil.get_path(self.root_dir, scene_name, first_frame_idx)
+        first_left_cam_pose = first_cam_info["cam_to_world"]
+        first_center_cam_pose = first_cam_info["cam_to_world_O"]
+        first_target_point_cloud = DataLoadUtil.load_from_preprocessed_pts(first_view_path)
+        first_pts_num = first_target_point_cloud.shape[0]
+        first_downsampled_target_point_cloud = PtsUtil.random_downsample_point_cloud(first_target_point_cloud, self.pts_num)
+        first_to_world_rot_6d = PoseUtil.matrix_to_rotation_6d_numpy(np.asarray(first_left_cam_pose[:3,:3]))
+        first_to_world_trans = first_left_cam_pose[:3,3]
+        first_to_world_9d = np.concatenate([first_to_world_rot_6d, first_to_world_trans], axis=0)
+        diag = DataLoadUtil.get_bbox_diag(self.model_dir, scene_name)
+        voxel_threshold = diag*0.02
+        first_O_to_first_L_pose = np.dot(np.linalg.inv(first_left_cam_pose), first_center_cam_pose)
+        scene_path = os.path.join(self.root_dir, scene_name)
+        model_points_normals = DataLoadUtil.load_points_normals(self.root_dir, scene_name)
+
+        data_item = {
+            "first_pts_num": np.asarray(
+                first_pts_num, dtype=np.int32
+            ),
+            "first_pts": np.asarray([first_downsampled_target_point_cloud],dtype=np.float32),
+            "combined_scanned_pts": np.asarray(first_downsampled_target_point_cloud,dtype=np.float32),
+            "first_to_world_9d": np.asarray([first_to_world_9d],dtype=np.float32),
+            "scene_name": scene_name,
+            "max_coverage_rate": max_coverage_rate,
+            "voxel_threshold": voxel_threshold,
+            "filter_degree": self.filter_degree,
+            "O_to_L_pose": first_O_to_first_L_pose,
+            "first_frame_coverage": first_frame_coverage,
+            "scene_path": scene_path,
+            "model_points_normals": model_points_normals,
+            "best_seq_len": data_item_info["best_seq_len"],
+            "first_frame_id": first_frame_idx,
+        }
+        return data_item
+
+    def __len__(self):
+        return len(self.datalist)
+    
+    def get_collate_fn(self):
+        def collate_fn(batch):
+            collate_data = {}
+            collate_data["first_pts"] = [torch.tensor(item['first_pts']) for item in batch]
+            collate_data["first_to_world_9d"] = [torch.tensor(item['first_to_world_9d']) for item in batch]
+            collate_data["combined_scanned_pts"] = torch.stack([torch.tensor(item['combined_scanned_pts']) for item in batch])
+            for key in batch[0].keys():
+                if key not in ["first_pts", "first_to_world_9d", "combined_scanned_pts"]:
+                    collate_data[key] = [item[key] for item in batch]
+            return collate_data
+        return collate_fn
+
+# -------------- Debug ---------------- #
+if __name__ == "__main__":
+    import torch
+    seed = 0
+    torch.manual_seed(seed)
+    np.random.seed(seed)
+    config = {
+        "root_dir": "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy",
+        "split_file": "/home/data/hofee/project/nbv_rec/data/OmniObject3d_train.txt",
+        "model_dir": "/home/data/hofee/project/nbv_rec/data/scaled_object_meshes",
+        "ratio": 0.005,
+        "batch_size": 2,
+        "filter_degree": 75,
+        "num_workers": 0,
+        "pts_num": 32684,
+        "type": namespace.Mode.TEST,
+        "load_from_preprocess": True
+    }
+    ds = SeqNBVReconstructionDataset(config)
+    print(len(ds))
+    #ds.__getitem__(10)
+    dl = ds.get_loader(shuffle=True)
+    for idx, data in enumerate(dl):
+        data = ds.process_batch(data, "cuda:0")
+        print(data)
+        # ------  Debug Start ------
+        import ipdb;ipdb.set_trace()
+        # ------  Debug End ------+

core/pipeline.py

@@ -0,0 +1,136 @@
+import torch
+import time
+from torch import nn
+import PytorchBoot.namespace as namespace
+import PytorchBoot.stereotype as stereotype
+from PytorchBoot.factory.component_factory import ComponentFactory
+from PytorchBoot.utils import Log
+
+
+@stereotype.pipeline("nbv_reconstruction_pipeline")
+class NBVReconstructionPipeline(nn.Module):
+    def __init__(self, config):
+        super(NBVReconstructionPipeline, self).__init__()
+        self.config = config
+        self.module_config = config["modules"]
+        
+        self.pts_encoder = ComponentFactory.create(
+            namespace.Stereotype.MODULE, self.module_config["pts_encoder"]
+        )
+        self.pose_encoder = ComponentFactory.create(
+            namespace.Stereotype.MODULE, self.module_config["pose_encoder"]
+        )
+        self.seq_encoder = ComponentFactory.create(
+            namespace.Stereotype.MODULE, self.module_config["seq_encoder"]
+        )
+        self.view_finder = ComponentFactory.create(
+            namespace.Stereotype.MODULE, self.module_config["view_finder"]
+        )
+        
+
+        self.eps = float(self.config["eps"])
+
+    def forward(self, data):
+        mode = data["mode"]
+
+        if mode == namespace.Mode.TRAIN:
+            return self.forward_train(data)
+        elif mode == namespace.Mode.TEST:
+            return self.forward_test(data)
+        else:
+            Log.error("Unknown mode: {}".format(mode), True)
+
+    def pertube_data(self, gt_delta_9d):
+        bs = gt_delta_9d.shape[0]
+        random_t = (
+            torch.rand(bs, device=gt_delta_9d.device) * (1.0 - self.eps) + self.eps
+        )
+        random_t = random_t.unsqueeze(-1)
+        mu, std = self.view_finder.marginal_prob(gt_delta_9d, random_t)
+        std = std.view(-1, 1)
+        z = torch.randn_like(gt_delta_9d)
+        perturbed_x = mu + z * std
+        target_score = -z * std / (std**2)
+        return perturbed_x, random_t, target_score, std
+
+    def forward_train(self, data):
+        main_feat = self.get_main_feat(data)
+        """ get std """
+        best_to_world_pose_9d_batch = data["best_to_world_pose_9d"]
+        perturbed_x, random_t, target_score, std = self.pertube_data(
+            best_to_world_pose_9d_batch
+        )
+        input_data = {
+            "sampled_pose": perturbed_x,
+            "t": random_t,
+            "main_feat": main_feat,
+        }
+        estimated_score = self.view_finder(input_data)
+        output = {
+            "estimated_score": estimated_score,
+            "target_score": target_score,
+            "std": std,
+        }
+        return output
+
+    def forward_test(self, data):
+        main_feat = self.get_main_feat(data)
+        estimated_delta_rot_9d, in_process_sample = self.view_finder.next_best_view(
+            main_feat
+        )
+        result = {
+            "pred_pose_9d": estimated_delta_rot_9d,
+            "in_process_sample": in_process_sample,
+        }
+        return result
+
+    def get_main_feat(self, data):
+        scanned_n_to_world_pose_9d_batch = data[
+            "scanned_n_to_world_pose_9d"
+        ]  # List(B): Tensor(S x 9)
+        scanned_pts_mask_batch = data["scanned_pts_mask"] # List(B): Tensor(S x N)
+
+        device = next(self.parameters()).device
+
+        embedding_list_batch = []
+
+        combined_scanned_pts_batch = data["combined_scanned_pts"]  # Tensor(B x N x 3)
+        global_scanned_feat, per_point_feat_batch = self.pts_encoder.encode_points(
+            combined_scanned_pts_batch, require_per_point_feat=True
+        )  # global_scanned_feat: Tensor(B x Dg)
+        batch_size = len(scanned_n_to_world_pose_9d_batch)
+        for i in range(batch_size):
+            seq_len = len(scanned_n_to_world_pose_9d_batch[i])
+            scanned_n_to_world_pose_9d = scanned_n_to_world_pose_9d_batch[i].to(device)  # Tensor(S x 9)
+            scanned_pts_mask = scanned_pts_mask_batch[i] # Tensor(S x N)
+            per_point_feat = per_point_feat_batch[i] # Tensor(N x Dp)
+            partial_point_feat_seq = []
+            for j in range(seq_len):
+                partial_per_point_feat = per_point_feat[scanned_pts_mask[j]]
+                if partial_per_point_feat.shape[0] == 0:
+                    partial_point_feat = torch.zeros(per_point_feat.shape[1], device=device)
+                else:
+                    partial_point_feat = torch.mean(partial_per_point_feat, dim=0) # Tensor(Dp)
+                partial_point_feat_seq.append(partial_point_feat)
+            partial_point_feat_seq = torch.stack(partial_point_feat_seq, dim=0) # Tensor(S x Dp)
+            
+            pose_feat_seq = self.pose_encoder.encode_pose(scanned_n_to_world_pose_9d)  # Tensor(S x Dp) 
+
+            seq_embedding = torch.cat([partial_point_feat_seq, pose_feat_seq], dim=-1)
+            
+            embedding_list_batch.append(seq_embedding) # List(B): Tensor(S x (Dp))
+        
+        seq_feat = self.seq_encoder.encode_sequence(embedding_list_batch) # Tensor(B x Ds)
+        main_feat = torch.cat([seq_feat, global_scanned_feat], dim=-1) # Tensor(B x (Ds+Dg))
+
+        if torch.isnan(main_feat).any():
+            for i in range(len(main_feat)):
+                if torch.isnan(main_feat[i]).any():
+                    scanned_pts_mask = scanned_pts_mask_batch[i]
+                    Log.info(f"scanned_pts_mask shape: {scanned_pts_mask.shape}")
+                    Log.info(f"scanned_pts_mask sum: {scanned_pts_mask.sum()}")
+                    import ipdb
+                    ipdb.set_trace()
+            Log.error("nan in main_feat", True)
+
+        return main_feat

core/seq_dataset.py

@@ -1,163 +1,204 @@
-import numpy as np
-from PytorchBoot.dataset import BaseDataset
-import PytorchBoot.namespace as namespace
-import PytorchBoot.stereotype as stereotype
-from PytorchBoot.utils.log_util import Log
-import torch
-import os
-import sys
-sys.path.append(r"/home/data/hofee/project/nbv_rec/nbv_reconstruction")
-
-from utils.data_load import DataLoadUtil
-from utils.pose import PoseUtil
-from utils.pts import PtsUtil
-
-@stereotype.dataset("seq_nbv_reconstruction_dataset")
-class SeqNBVReconstructionDataset(BaseDataset):
-    def __init__(self, config):
-        super(SeqNBVReconstructionDataset, self).__init__(config)
-        self.type = config["type"]
-        if self.type != namespace.Mode.TEST:
-            Log.error("Dataset <seq_nbv_reconstruction_dataset> Only support test mode", terminate=True)
-        self.config = config
-        self.root_dir = config["root_dir"]
-        self.split_file_path = config["split_file"]
-        self.scene_name_list = self.load_scene_name_list()
-        self.datalist = self.get_datalist()
-        self.pts_num = config["pts_num"]
-        
-        self.model_dir = config["model_dir"]
-        self.filter_degree = config["filter_degree"]
-        self.load_from_preprocess = config.get("load_from_preprocess", False)
-        
-
-    def load_scene_name_list(self):
-        scene_name_list = []
-        with open(self.split_file_path, "r") as f:
-            for line in f:
-                scene_name = line.strip()
-                scene_name_list.append(scene_name)
-        return scene_name_list
-    
-    def get_datalist(self):
-        datalist = []
-        for scene_name in self.scene_name_list:
-            seq_num = DataLoadUtil.get_label_num(self.root_dir, scene_name)
-            scene_max_coverage_rate = 0
-            scene_max_cr_idx = 0
-
-            for seq_idx in range(seq_num):
-                label_path = DataLoadUtil.get_label_path(self.root_dir, scene_name, seq_idx)
-                label_data = DataLoadUtil.load_label(label_path)
-                max_coverage_rate = label_data["max_coverage_rate"]
-                if max_coverage_rate > scene_max_coverage_rate:
-                    scene_max_coverage_rate = max_coverage_rate
-                    scene_max_cr_idx = seq_idx
-        
-            label_path = DataLoadUtil.get_label_path(self.root_dir, scene_name, scene_max_cr_idx)
-            label_data = DataLoadUtil.load_label(label_path)
-            first_frame = label_data["best_sequence"][0]
-            best_seq_len = len(label_data["best_sequence"])
-            datalist.append({
-                    "scene_name": scene_name,
-                    "first_frame": first_frame,
-                    "max_coverage_rate": scene_max_coverage_rate,
-                    "best_seq_len": best_seq_len,
-                    "label_idx": scene_max_cr_idx,
-                })
-        return datalist
-
-    def __getitem__(self, index):
-        data_item_info = self.datalist[index]
-        first_frame_idx = data_item_info["first_frame"][0]
-        first_frame_coverage = data_item_info["first_frame"][1]
-        max_coverage_rate = data_item_info["max_coverage_rate"]
-        scene_name = data_item_info["scene_name"]
-        first_cam_info = DataLoadUtil.load_cam_info(DataLoadUtil.get_path(self.root_dir, scene_name, first_frame_idx), binocular=True)
-       
-        first_view_path = DataLoadUtil.get_path(self.root_dir, scene_name, first_frame_idx)
-        first_left_cam_pose = first_cam_info["cam_to_world"]
-        first_right_cam_pose = first_cam_info["cam_to_world_R"]
-        first_center_cam_pose = first_cam_info["cam_to_world_O"]
-        if self.load_from_preprocess:
-            first_downsampled_target_point_cloud = DataLoadUtil.load_from_preprocessed_pts(first_view_path)
-        else:    
-            first_depth_L, first_depth_R = DataLoadUtil.load_depth(first_view_path, first_cam_info['near_plane'], first_cam_info['far_plane'], binocular=True)
-            
-            first_point_cloud_L = DataLoadUtil.get_point_cloud(first_depth_L, first_cam_info['cam_intrinsic'], first_left_cam_pose)['points_world']
-            first_point_cloud_R = DataLoadUtil.get_point_cloud(first_depth_R, first_cam_info['cam_intrinsic'], first_right_cam_pose)['points_world']
-            
-            first_point_cloud_L = PtsUtil.random_downsample_point_cloud(first_point_cloud_L, 65536)
-            first_point_cloud_R = PtsUtil.random_downsample_point_cloud(first_point_cloud_R, 65536)
-            first_overlap_points = PtsUtil.get_overlapping_points(first_point_cloud_L, first_point_cloud_R)
-            first_downsampled_target_point_cloud = PtsUtil.random_downsample_point_cloud(first_overlap_points, self.pts_num)
-     
-        first_to_world_rot_6d = PoseUtil.matrix_to_rotation_6d_numpy(np.asarray(first_left_cam_pose[:3,:3]))
-        first_to_world_trans = first_left_cam_pose[:3,3]
-        first_to_world_9d = np.concatenate([first_to_world_rot_6d, first_to_world_trans], axis=0)
-        diag = DataLoadUtil.get_bbox_diag(self.model_dir, scene_name)
-        voxel_threshold = diag*0.02
-        first_O_to_first_L_pose = np.dot(np.linalg.inv(first_left_cam_pose), first_center_cam_pose)
-        scene_path = os.path.join(self.root_dir, scene_name)
-        model_points_normals = DataLoadUtil.load_points_normals(self.root_dir, scene_name)
-
-        data_item = {
-            "first_pts": np.asarray([first_downsampled_target_point_cloud],dtype=np.float32),
-            "combined_scanned_pts": np.asarray(first_downsampled_target_point_cloud,dtype=np.float32),
-            "first_to_world_9d": np.asarray([first_to_world_9d],dtype=np.float32),
-            "scene_name": scene_name,
-            "max_coverage_rate": max_coverage_rate,
-            "voxel_threshold": voxel_threshold,
-            "filter_degree": self.filter_degree,
-            "O_to_L_pose": first_O_to_first_L_pose,
-            "first_frame_coverage": first_frame_coverage,
-            "scene_path": scene_path,
-            "model_points_normals": model_points_normals,
-            "best_seq_len": data_item_info["best_seq_len"],
-            "first_frame_id": first_frame_idx,
-        }
-        return data_item
-
-    def __len__(self):
-        return len(self.datalist)
-    
-    def get_collate_fn(self):
-        def collate_fn(batch):
-            collate_data = {}
-            collate_data["first_pts"] = [torch.tensor(item['first_pts']) for item in batch]
-            collate_data["first_to_world_9d"] = [torch.tensor(item['first_to_world_9d']) for item in batch]
-            collate_data["combined_scanned_pts"] = torch.stack([torch.tensor(item['combined_scanned_pts']) for item in batch])
-            for key in batch[0].keys():
-                if key not in ["first_pts", "first_to_world_9d", "combined_scanned_pts"]:
-                    collate_data[key] = [item[key] for item in batch]
-            return collate_data
-        return collate_fn
-
-# -------------- Debug ---------------- #
-if __name__ == "__main__":
-    import torch
-    seed = 0
-    torch.manual_seed(seed)
-    np.random.seed(seed)
-    config = {
-        "root_dir": "/home/data/hofee/project/nbv_rec/data/nbv_rec_data_512_preproc_npy",
-        "split_file": "/home/data/hofee/project/nbv_rec/data/OmniObject3d_train.txt",
-        "model_dir": "/home/data/hofee/project/nbv_rec/data/scaled_object_meshes",
-        "ratio": 0.005,
-        "batch_size": 2,
-        "filter_degree": 75,
-        "num_workers": 0,
-        "pts_num": 32684,
-        "type": namespace.Mode.TEST,
-        "load_from_preprocess": True
-    }
-    ds = SeqNBVReconstructionDataset(config)
-    print(len(ds))
-    #ds.__getitem__(10)
-    dl = ds.get_loader(shuffle=True)
-    for idx, data in enumerate(dl):
-        data = ds.process_batch(data, "cuda:0")
-        print(data)
-        # ------  Debug Start ------
-        import ipdb;ipdb.set_trace()
-        # ------  Debug End ------+
+import numpy as np
+from PytorchBoot.dataset import BaseDataset
+import PytorchBoot.namespace as namespace
+import PytorchBoot.stereotype as stereotype
+from PytorchBoot.config import ConfigManager
+from PytorchBoot.utils.log_util import Log
+import torch
+import os
+import sys
+
+sys.path.append(r"/media/hofee/data/project/python/nbv_reconstruction/nbv_reconstruction")
+
+from utils.data_load import DataLoadUtil
+from utils.pose import PoseUtil
+from utils.pts import PtsUtil
+
+
+@stereotype.dataset("seq_reconstruction_dataset")
+class SeqReconstructionDataset(BaseDataset):
+    def __init__(self, config):
+        super(SeqReconstructionDataset, self).__init__(config)
+        self.config = config
+        self.root_dir = config["root_dir"]
+        self.split_file_path = config["split_file"]
+        self.scene_name_list = self.load_scene_name_list()
+        self.datalist = self.get_datalist()
+
+        self.pts_num = config["pts_num"]
+        self.type = config["type"]
+        self.cache = config.get("cache")
+        self.load_from_preprocess = config.get("load_from_preprocess", False)
+
+        if self.type == namespace.Mode.TEST:
+            #self.model_dir = config["model_dir"]
+            self.filter_degree = config["filter_degree"]
+        if self.type == namespace.Mode.TRAIN:
+            scale_ratio = 1
+            self.datalist = self.datalist*scale_ratio
+        if self.cache:
+            expr_root = ConfigManager.get("runner", "experiment", "root_dir")
+            expr_name = ConfigManager.get("runner", "experiment", "name")
+            self.cache_dir = os.path.join(expr_root, expr_name, "cache")
+            # self.preprocess_cache()
+
+    def load_scene_name_list(self):
+        scene_name_list = []
+        with open(self.split_file_path, "r") as f:
+            for line in f:
+                scene_name = line.strip()
+                if not os.path.exists(os.path.join(self.root_dir, scene_name)):
+                    continue
+                scene_name_list.append(scene_name)
+        return scene_name_list
+    
+    def get_scene_name_list(self):
+        return self.scene_name_list
+
+    def get_datalist(self):
+        datalist = []
+        total = len(self.scene_name_list)
+        for idx, scene_name in enumerate(self.scene_name_list):
+            print(f"processing {scene_name} ({idx}/{total})")
+            scene_max_cr_idx = 0
+            frame_len = DataLoadUtil.get_scene_seq_length(self.root_dir, scene_name)
+
+            for i in range(frame_len):
+                path = DataLoadUtil.get_path(self.root_dir, scene_name, i)
+                pts = DataLoadUtil.load_from_preprocessed_pts(path, "npy")
+                if pts.shape[0] == 0:
+                    continue
+            datalist.append({
+                "scene_name": scene_name,
+                "first_frame": i,
+                "best_seq_len": -1,
+                "max_coverage_rate": 1.0,
+                "label_idx": scene_max_cr_idx,
+            })  
+        return datalist
+
+    def preprocess_cache(self):
+        Log.info("preprocessing cache...")
+        for item_idx in range(len(self.datalist)):
+            self.__getitem__(item_idx)
+        Log.success("finish preprocessing cache.")
+
+    def load_from_cache(self, scene_name, curr_frame_idx):
+        cache_name = f"{scene_name}_{curr_frame_idx}.txt"
+        cache_path = os.path.join(self.cache_dir, cache_name)
+        if os.path.exists(cache_path):
+            data = np.loadtxt(cache_path)
+            return data
+        else:
+            return None
+
+    def save_to_cache(self, scene_name, curr_frame_idx, data):
+        cache_name = f"{scene_name}_{curr_frame_idx}.txt"
+        cache_path = os.path.join(self.cache_dir, cache_name)
+        try:
+            np.savetxt(cache_path, data)
+        except Exception as e:
+            Log.error(f"Save cache failed: {e}")
+
+    def seq_combined_pts(self, scene, frame_idx_list):
+        all_combined_pts = []   
+        for i in frame_idx_list:
+            path = DataLoadUtil.get_path(self.root_dir, scene, i)
+            pts = DataLoadUtil.load_from_preprocessed_pts(path,"npy")
+            if pts.shape[0] == 0:
+                continue
+            all_combined_pts.append(pts)
+        all_combined_pts = np.vstack(all_combined_pts)
+        downsampled_all_pts = PtsUtil.voxel_downsample_point_cloud(all_combined_pts, 0.003)
+        return downsampled_all_pts
+
+    def __getitem__(self, index):
+        data_item_info = self.datalist[index]
+        max_coverage_rate = data_item_info["max_coverage_rate"]
+        best_seq_len = data_item_info["best_seq_len"]
+        scene_name = data_item_info["scene_name"]
+        (
+            scanned_views_pts,
+            scanned_coverages_rate,
+            scanned_n_to_world_pose,
+        ) = ([], [], [])
+        view = data_item_info["first_frame"]
+        frame_idx = view
+        view_path = DataLoadUtil.get_path(self.root_dir, scene_name, frame_idx)
+        cam_info = DataLoadUtil.load_cam_info(view_path, binocular=True)
+        
+        n_to_world_pose = cam_info["cam_to_world"]  
+        target_point_cloud = (
+            DataLoadUtil.load_from_preprocessed_pts(view_path)
+        )
+        downsampled_target_point_cloud = PtsUtil.random_downsample_point_cloud(
+            target_point_cloud, self.pts_num
+        )
+        scanned_views_pts.append(downsampled_target_point_cloud)
+        
+        n_to_world_6d = PoseUtil.matrix_to_rotation_6d_numpy(
+            np.asarray(n_to_world_pose[:3, :3])
+        ) 
+        first_left_cam_pose = cam_info["cam_to_world"]
+        first_center_cam_pose = cam_info["cam_to_world_O"]
+        first_O_to_first_L_pose = np.dot(np.linalg.inv(first_left_cam_pose), first_center_cam_pose)
+        n_to_world_trans = n_to_world_pose[:3, 3]
+        n_to_world_9d = np.concatenate([n_to_world_6d, n_to_world_trans], axis=0)
+        scanned_n_to_world_pose.append(n_to_world_9d)
+
+        frame_list = []
+        for i in range(DataLoadUtil.get_scene_seq_length(self.root_dir, scene_name)):
+            frame_list.append(i)
+        gt_pts = self.seq_combined_pts(scene_name, frame_list)
+        data_item = {
+            "first_scanned_pts": np.asarray(scanned_views_pts, dtype=np.float32), # Ndarray(S x Nv x 3)
+            "first_scanned_n_to_world_pose_9d": np.asarray(scanned_n_to_world_pose, dtype=np.float32), # Ndarray(S x 9)
+            "seq_max_coverage_rate": max_coverage_rate, # Float, range(0, 1)
+            "best_seq_len": best_seq_len, # Int
+            "scene_name": scene_name, # String
+            "gt_pts": gt_pts, # Ndarray(N x 3)
+            "scene_path": os.path.join(self.root_dir, scene_name), # String
+            "O_to_L_pose": first_O_to_first_L_pose,
+        }
+
+        return data_item
+
+    def __len__(self):
+        return len(self.datalist)
+
+
+# -------------- Debug ---------------- #
+if __name__ == "__main__":
+    import torch
+    from tqdm import tqdm
+    import pickle
+    import os
+
+    seed = 0
+    torch.manual_seed(seed)
+    np.random.seed(seed)
+    
+    config = {
+        "root_dir": "/media/hofee/data/data/new_testset",
+        "source": "seq_reconstruction_dataset",
+        "split_file": "/media/hofee/data/data/OmniObject3d_test.txt",
+        "load_from_preprocess": True,
+        "filter_degree": 75,
+        "num_workers": 0,
+        "pts_num": 8192,
+        "type": namespace.Mode.TEST,
+    }
+    
+    output_dir = "/media/hofee/data/data/new_testset_output"
+    os.makedirs(output_dir, exist_ok=True)
+
+    ds = SeqReconstructionDataset(config)
+    for i in tqdm(range(len(ds)), desc="processing dataset"):
+        output_path = os.path.join(output_dir, f"item_{i}.pkl")
+        item = ds.__getitem__(i)
+        for key, value in item.items():
+            if isinstance(value, np.ndarray):
+                item[key] = value.tolist()
+        #import ipdb; ipdb.set_trace()
+        with open(output_path, "wb") as f:
+            pickle.dump(item, f)

core/seq_dataset_preprocessed.py

@@ -0,0 +1,82 @@
+import numpy as np
+from PytorchBoot.dataset import BaseDataset
+import PytorchBoot.namespace as namespace
+import PytorchBoot.stereotype as stereotype
+from PytorchBoot.config import ConfigManager
+from PytorchBoot.utils.log_util import Log
+import pickle
+import torch
+import os
+import sys
+
+sys.path.append(r"C:\Document\Local Project\nbv_rec\nbv_reconstruction")
+
+from utils.data_load import DataLoadUtil
+from utils.pose import PoseUtil
+from utils.pts import PtsUtil
+
+@stereotype.dataset("seq_reconstruction_dataset_preprocessed")
+class SeqReconstructionDatasetPreprocessed(BaseDataset):
+    def __init__(self, config):
+        super(SeqReconstructionDatasetPreprocessed, self).__init__(config)
+        self.config = config
+        self.root_dir = config["root_dir"]
+        self.real_root_dir = r"/media/hofee/data/data/new_testset"
+        self.item_list = os.listdir(self.root_dir)
+
+    def __getitem__(self, index):
+        data = pickle.load(open(os.path.join(self.root_dir, self.item_list[index]), "rb"))
+        data_item = {
+            "first_scanned_pts": np.asarray(data["first_scanned_pts"], dtype=np.float32), # Ndarray(S x Nv x 3)
+            "first_scanned_n_to_world_pose_9d": np.asarray(data["first_scanned_n_to_world_pose_9d"], dtype=np.float32), # Ndarray(S x 9)
+            "seq_max_coverage_rate": data["seq_max_coverage_rate"], # Float, range(0, 1)
+            "best_seq_len": data["best_seq_len"], # Int
+            "scene_name": data["scene_name"], # String
+            "gt_pts": np.asarray(data["gt_pts"], dtype=np.float32), # Ndarray(N x 3)
+            "scene_path": os.path.join(self.real_root_dir, data["scene_name"]), # String
+            "O_to_L_pose": np.asarray(data["O_to_L_pose"], dtype=np.float32),
+        }
+        return data_item
+
+    def __len__(self):
+        return len(self.item_list)
+
+# -------------- Debug ---------------- #
+if __name__ == "__main__":
+    import torch
+
+    seed = 0
+    torch.manual_seed(seed)
+    np.random.seed(seed)
+    '''
+    OmniObject3d_test:
+    root_dir: "H:\\AI\\Datasets\\packed_test_data"
+    model_dir: "H:\\AI\\Datasets\\scaled_object_meshes"
+    source: seq_reconstruction_dataset
+    split_file: "H:\\AI\\Datasets\\data_list\\OmniObject3d_test.txt"
+    type: test
+    filter_degree: 75
+    eval_list:
+      - pose_diff
+      - coverage_rate_increase
+    ratio: 0.1
+    batch_size: 1
+    num_workers: 12
+    pts_num: 8192
+    load_from_preprocess: True
+    '''
+    config = {
+        "root_dir": "H:\\AI\\Datasets\\packed_test_data",
+        "source": "seq_reconstruction_dataset",
+        "split_file": "H:\\AI\\Datasets\\data_list\\OmniObject3d_test.txt",
+        "load_from_preprocess": True,
+        "ratio": 1,
+        "filter_degree": 75,
+        "num_workers": 0,
+        "pts_num": 8192,
+        "type": "test",
+    }
+    ds = SeqReconstructionDataset(config)
+    print(len(ds))
+    print(ds.__getitem__(10))
+    

preprocess/clean_preprocessed_data.py

@@ -0,0 +1,43 @@
+import os
+import shutil
+
+def clean_scene_data(root, scene):
+    # 清理目标点云数据
+    pts_dir = os.path.join(root, scene, "pts")
+    if os.path.exists(pts_dir):
+        shutil.rmtree(pts_dir)
+        print(f"已删除 {pts_dir}")
+
+    # 清理法线数据
+    nrm_dir = os.path.join(root, scene, "nrm")
+    if os.path.exists(nrm_dir):
+        shutil.rmtree(nrm_dir)
+        print(f"已删除 {nrm_dir}")
+
+    # 清理扫描点索引数据
+    scan_points_indices_dir = os.path.join(root, scene, "scan_points_indices")
+    if os.path.exists(scan_points_indices_dir):
+        shutil.rmtree(scan_points_indices_dir)
+        print(f"已删除 {scan_points_indices_dir}")
+
+    # 删除扫描点数据文件
+    scan_points_file = os.path.join(root, scene, "scan_points.txt")
+    if os.path.exists(scan_points_file):
+        os.remove(scan_points_file)
+        print(f"已删除 {scan_points_file}")
+
+def clean_all_scenes(root, scene_list):
+    for idx, scene in enumerate(scene_list):
+        print(f"正在清理场景 {scene} ({idx+1}/{len(scene_list)})")
+        clean_scene_data(root, scene)
+
+if __name__ == "__main__":
+    root = r"c:\Document\Local Project\nbv_rec\nbv_reconstruction\temp"
+    scene_list = os.listdir(root)
+    from_idx = 0
+    to_idx = len(scene_list)
+    print(f"正在清理场景 {scene_list[from_idx:to_idx]}")
+
+    clean_all_scenes(root, scene_list[from_idx:to_idx])
+    print("清理完成")
+

preprocess/pack_preprocessed_data.py

@@ -0,0 +1,48 @@
+import os
+import shutil
+
+def pack_scene_data(root, scene, output_dir):
+    scene_dir = os.path.join(output_dir, scene)
+    if not os.path.exists(scene_dir):
+        os.makedirs(scene_dir)
+
+    pts_dir = os.path.join(root, scene, "pts")
+    if os.path.exists(pts_dir):
+        shutil.move(pts_dir, os.path.join(scene_dir, "pts"))
+        
+    scan_points_indices_dir = os.path.join(root, scene, "scan_points_indices")
+    if os.path.exists(scan_points_indices_dir):
+        shutil.move(scan_points_indices_dir, os.path.join(scene_dir, "scan_points_indices")) 
+
+    scan_points_file = os.path.join(root, scene, "scan_points.txt")
+    if os.path.exists(scan_points_file):
+        shutil.move(scan_points_file, os.path.join(scene_dir, "scan_points.txt"))
+
+    model_pts_nrm_file = os.path.join(root, scene, "points_and_normals.txt")
+    if os.path.exists(model_pts_nrm_file):
+        shutil.move(model_pts_nrm_file, os.path.join(scene_dir, "points_and_normals.txt"))
+
+    camera_dir = os.path.join(root, scene, "camera_params")
+    if os.path.exists(camera_dir):
+        shutil.move(camera_dir, os.path.join(scene_dir, "camera_params"))
+
+    scene_info_file = os.path.join(root, scene, "scene_info.json")
+    if os.path.exists(scene_info_file):
+        shutil.move(scene_info_file, os.path.join(scene_dir, "scene_info.json"))
+
+def pack_all_scenes(root, scene_list, output_dir):
+    for idx, scene in enumerate(scene_list):
+        print(f"正在打包场景 {scene} ({idx+1}/{len(scene_list)})")
+        pack_scene_data(root, scene, output_dir)
+
+if __name__ == "__main__":
+    root = r"H:\AI\Datasets\nbv_rec_part2"
+    output_dir = r"H:\AI\Datasets\scene_info_part2"
+    scene_list = os.listdir(root)
+    from_idx = 0
+    to_idx = len(scene_list)
+    print(f"正在打包场景 {scene_list[from_idx:to_idx]}")
+
+    pack_all_scenes(root, scene_list[from_idx:to_idx], output_dir)
+    print("打包完成")
+

preprocess/pack_upload_data.py

@@ -0,0 +1,41 @@
+import os
+import shutil
+
+def pack_scene_data(root, scene, output_dir):
+    scene_dir = os.path.join(output_dir, scene)
+    if not os.path.exists(scene_dir):
+        os.makedirs(scene_dir)
+
+    pts_dir = os.path.join(root, scene, "pts")
+    if os.path.exists(pts_dir):
+        shutil.move(pts_dir, os.path.join(scene_dir, "pts"))
+
+    camera_dir = os.path.join(root, scene, "camera_params")
+    if os.path.exists(camera_dir):
+        shutil.move(camera_dir, os.path.join(scene_dir, "camera_params"))
+
+    scene_info_file = os.path.join(root, scene, "scene_info.json")
+    if os.path.exists(scene_info_file):
+        shutil.move(scene_info_file, os.path.join(scene_dir, "scene_info.json"))
+
+    label_dir = os.path.join(root, scene, "label")
+    if os.path.exists(label_dir):
+        shutil.move(label_dir, os.path.join(scene_dir, "label"))
+
+
+def pack_all_scenes(root, scene_list, output_dir):
+    for idx, scene in enumerate(scene_list):
+        print(f"packing {scene} ({idx+1}/{len(scene_list)})")
+        pack_scene_data(root, scene, output_dir)
+
+if __name__ == "__main__":
+    root = r"H:\AI\Datasets\nbv_rec_part2"
+    output_dir = r"H:\AI\Datasets\upload_part2"
+    scene_list = os.listdir(root)
+    from_idx = 0
+    to_idx = len(scene_list)
+    print(f"packing {scene_list[from_idx:to_idx]}")
+
+    pack_all_scenes(root, scene_list[from_idx:to_idx], output_dir)
+    print("packing done")
+

preprocess/preprocessor.py

@@ -15,11 +15,18 @@ def save_np_pts(path, pts: np.ndarray, file_type="txt"):
     else:
         np.save(path, pts)
 
+
 def save_target_points(root, scene, frame_idx, target_points: np.ndarray, file_type="txt"):
     pts_path = os.path.join(root,scene, "pts", f"{frame_idx}.{file_type}")
     if not os.path.exists(os.path.join(root,scene, "pts")):
         os.makedirs(os.path.join(root,scene, "pts"))
     save_np_pts(pts_path, target_points, file_type)
+
+def save_target_normals(root, scene, frame_idx, target_normals: np.ndarray, file_type="txt"):
+    pts_path = os.path.join(root,scene, "nrm", f"{frame_idx}.{file_type}")
+    if not os.path.exists(os.path.join(root,scene, "nrm")):
+        os.makedirs(os.path.join(root,scene, "nrm"))
+    save_np_pts(pts_path, target_normals, file_type)
     
 def save_scan_points_indices(root, scene, frame_idx, scan_points_indices: np.ndarray, file_type="txt"):
     indices_path = os.path.join(root,scene, "scan_points_indices", f"{frame_idx}.{file_type}")
@@ -30,35 +37,42 @@ def save_scan_points_indices(root, scene, frame_idx, scan_points_indices: np.nda
 def save_scan_points(root, scene, scan_points: np.ndarray):
     scan_points_path = os.path.join(root,scene, "scan_points.txt")
     save_np_pts(scan_points_path, scan_points)
-    
 
-def old_get_world_points(depth, cam_intrinsic, cam_extrinsic):
-    h, w = depth.shape
-    i, j = np.meshgrid(np.arange(w), np.arange(h), indexing="xy")
-    # ----- Debug Trace ----- #
-    import ipdb; ipdb.set_trace()
-    # ------------------------ #
-    z = depth
-    x = (i - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
-    y = (j - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
-    points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
-    points_camera_aug = np.concatenate((points_camera, np.ones((points_camera.shape[0], 1))), axis=-1)
-    points_camera_world = np.dot(cam_extrinsic, points_camera_aug.T).T[:, :3]
-    
-    return points_camera_world
-
-def get_world_points(depth, mask, cam_intrinsic, cam_extrinsic):
+def get_world_points(depth, mask, cam_intrinsic, cam_extrinsic, random_downsample_N):
     z = depth[mask]
     i, j = np.nonzero(mask)
     x = (j - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
     y = (i - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
 
     points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
-    points_camera_aug = np.concatenate((points_camera, np.ones((points_camera.shape[0], 1))), axis=-1)
+    sampled_target_points = PtsUtil.random_downsample_point_cloud(
+                points_camera, random_downsample_N
+            )
+    points_camera_aug = np.concatenate((sampled_target_points, np.ones((sampled_target_points.shape[0], 1))), axis=-1)
     points_camera_world = np.dot(cam_extrinsic, points_camera_aug.T).T[:, :3]
     
     return points_camera_world
 
+def get_world_points_and_normal(depth, mask, normal, cam_intrinsic, cam_extrinsic, random_downsample_N):
+    z = depth[mask]
+    i, j = np.nonzero(mask)
+    x = (j - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
+    y = (i - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
+
+    points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
+    normal_camera = normal[mask].reshape(-1, 3)
+    sampled_target_points, idx = PtsUtil.random_downsample_point_cloud(
+                points_camera, random_downsample_N, require_idx=True
+            )
+    if len(sampled_target_points) == 0:
+        return np.zeros((0, 3)), np.zeros((0, 3))
+    sampled_normal_camera  = normal_camera[idx]
+
+    points_camera_aug = np.concatenate((sampled_target_points, np.ones((sampled_target_points.shape[0], 1))), axis=-1)
+    points_camera_world = np.dot(cam_extrinsic, points_camera_aug.T).T[:, :3]
+    
+    return points_camera_world, sampled_normal_camera
+
 def get_scan_points_indices(scan_points, mask, display_table_mask_label, cam_intrinsic, cam_extrinsic):
     scan_points_homogeneous = np.hstack((scan_points, np.ones((scan_points.shape[0], 1))))
     points_camera = np.dot(np.linalg.inv(cam_extrinsic), scan_points_homogeneous.T).T[:, :3]
@@ -74,16 +88,16 @@ def get_scan_points_indices(scan_points, mask, display_table_mask_label, cam_int
     return selected_points_indices
     
 
-def save_scene_data(root, scene, scene_idx=0, scene_total=1):
+def save_scene_data(root, scene, scene_idx=0, scene_total=1,file_type="txt"):
     
     ''' configuration '''
-    target_mask_label = (0, 255, 0, 255)
-    display_table_mask_label=(0, 0, 255, 255)
+    target_mask_label = (0, 255, 0)
+    display_table_mask_label=(0, 0, 255)
     random_downsample_N = 32768
-    voxel_size=0.002
+    voxel_size=0.003
     filter_degree = 75
     min_z = 0.2
-    max_z = 0.45
+    max_z = 0.5
     
     ''' scan points '''
     display_table_info = DataLoadUtil.get_display_table_info(root, scene)
@@ -103,7 +117,7 @@ def save_scene_data(root, scene, scene_idx=0, scene_total=1):
                 binocular=True
             )
         mask_L, mask_R = DataLoadUtil.load_seg(path, binocular=True)
-        
+        normal_L = DataLoadUtil.load_normal(path, binocular=True, left_only=True)
         ''' target points '''
         mask_img_L = mask_L
         mask_img_R = mask_R
@@ -112,28 +126,23 @@ def save_scene_data(root, scene, scene_idx=0, scene_total=1):
         target_mask_img_R = (mask_R == target_mask_label).all(axis=-1)
 
         
-        target_points_L = get_world_points(depth_L, target_mask_img_L, cam_info["cam_intrinsic"], cam_info["cam_to_world"])
-        target_points_R = get_world_points(depth_R, target_mask_img_R, cam_info["cam_intrinsic"], cam_info["cam_to_world_R"])
+        sampled_target_points_L, sampled_target_normal_L = get_world_points_and_normal(depth_L,target_mask_img_L,normal_L, cam_info["cam_intrinsic"], cam_info["cam_to_world"], random_downsample_N)
+        sampled_target_points_R = get_world_points(depth_R, target_mask_img_R, cam_info["cam_intrinsic"], cam_info["cam_to_world_R"], random_downsample_N)
 
-        sampled_target_points_L = PtsUtil.random_downsample_point_cloud(
-                target_points_L, random_downsample_N
-            )
-        sampled_target_points_R = PtsUtil.random_downsample_point_cloud(
-                target_points_R, random_downsample_N
-            )
 
         has_points = sampled_target_points_L.shape[0] > 0 and sampled_target_points_R.shape[0] > 0
         if has_points:
-            target_points = PtsUtil.get_overlapping_points(
-                    sampled_target_points_L, sampled_target_points_R, voxel_size
+            target_points, overlap_idx = PtsUtil.get_overlapping_points(
+                    sampled_target_points_L, sampled_target_points_R, voxel_size, require_idx=True
                 )
+            sampled_target_normal_L = sampled_target_normal_L[overlap_idx]
 
-        
-        has_points = target_points.shape[0] > 0
         if has_points:
-            points_normals = DataLoadUtil.load_points_normals(root, scene, display_table_as_world_space_origin=True)
-            target_points = PtsUtil.filter_points(
-                target_points, points_normals, cam_info["cam_to_world"],voxel_size=0.002, theta = filter_degree, z_range=(min_z, max_z)
+            has_points = target_points.shape[0] > 0
+
+        if has_points:
+            target_points, target_normals = PtsUtil.filter_points(
+                target_points, sampled_target_normal_L, cam_info["cam_to_world"], theta_limit = filter_degree, z_range=(min_z, max_z)
                 )
 
 
@@ -144,31 +153,33 @@ def save_scene_data(root, scene, scene_idx=0, scene_total=1):
         
         if not has_points:
             target_points = np.zeros((0, 3))
+            target_normals = np.zeros((0, 3))
             
-        save_target_points(root, scene, frame_id, target_points)
-        save_scan_points_indices(root, scene, frame_id, scan_points_indices)
+        save_target_points(root, scene, frame_id, target_points, file_type=file_type)
+        save_target_normals(root, scene, frame_id, target_normals, file_type=file_type)
+        save_scan_points_indices(root, scene, frame_id, scan_points_indices, file_type=file_type)
     
     save_scan_points(root, scene, scan_points) # The "done" flag of scene preprocess
 
 
 if __name__ == "__main__":
     #root = "/media/hofee/repository/new_data_with_normal"
-    root = r"C:\Document\Local Project\nbv_rec\nbv_reconstruction\test\test_sample"
-    list_path = r"C:\Document\Local Project\nbv_rec\nbv_reconstruction\test\test_sample/test_sample_list.txt"
-    scene_list = []
-    
-    with open(list_path, "r") as f:
-        for line in f:
-            scene_list.append(line.strip())
+    root = r"H:\AI\Datasets\nbv_rec_part2"
+    scene_list = os.listdir(root)
+    from_idx = 0 # 1000
+    to_idx = 600 # 1500
+
 
-    from_idx = 0
-    to_idx = len(scene_list)
     cnt = 0
     import time
     total = to_idx - from_idx
     for scene in scene_list[from_idx:to_idx]:
         start = time.time()
-        save_scene_data(root, scene, cnt, total)
+        if os.path.exists(os.path.join(root, scene, "scan_points.txt")):
+            print(f"Scene {scene} has been processed")
+            cnt+=1
+            continue
+        save_scene_data(root, scene, cnt, total, file_type="npy")
         cnt+=1
         end = time.time()
-        print(f"Time cost: {end-start}")
+        print(f"Time cost: {end-start}")

runners/heuristic.py

@@ -0,0 +1,425 @@
+import os
+import json
+from utils.render import RenderUtil
+from utils.pose import PoseUtil
+from utils.pts import PtsUtil
+from utils.reconstruction import ReconstructionUtil
+
+import torch
+from tqdm import tqdm
+import numpy as np
+import pickle
+
+from PytorchBoot.config import ConfigManager
+import PytorchBoot.namespace as namespace
+import PytorchBoot.stereotype as stereotype
+from PytorchBoot.factory import ComponentFactory
+
+from PytorchBoot.dataset import BaseDataset
+from PytorchBoot.runners.runner import Runner
+from PytorchBoot.utils import Log
+from PytorchBoot.status import status_manager
+from utils.data_load import DataLoadUtil
+
+@stereotype.runner("heuristic")
+class Heuristic(Runner):
+    def __init__(self, config_path):
+        
+        super().__init__(config_path)
+    
+        self.script_path = ConfigManager.get(namespace.Stereotype.RUNNER, "blender_script_path")
+        self.output_dir = ConfigManager.get(namespace.Stereotype.RUNNER, "output_dir")
+        self.voxel_size = ConfigManager.get(namespace.Stereotype.RUNNER, "voxel_size")
+        self.min_new_area = ConfigManager.get(namespace.Stereotype.RUNNER, "min_new_area")
+        self.heuristic_method = ConfigManager.get(namespace.Stereotype.RUNNER, "heuristic_method")
+        self.heuristic_method_config = ConfigManager.get("heuristic_methods", self.heuristic_method)
+        CM = 0.01
+        self.min_new_pts_num = self.min_new_area * (CM / self.voxel_size) **2
+
+        ''' Experiment '''
+        self.load_experiment("nbv_evaluator")
+        self.stat_result_path = os.path.join(self.output_dir, "stat.json")
+        if os.path.exists(self.stat_result_path):
+            with open(self.stat_result_path, "r") as f:
+                self.stat_result = json.load(f)
+        else:
+            self.stat_result = {}
+
+        ''' Test '''
+        self.test_config = ConfigManager.get(namespace.Stereotype.RUNNER, namespace.Mode.TEST)
+        self.test_dataset_name_list = self.test_config["dataset_list"]
+        self.test_set_list = []
+        self.test_writer_list = []
+        seen_name = set()
+        for test_dataset_name in self.test_dataset_name_list:
+            if test_dataset_name not in seen_name:
+                seen_name.add(test_dataset_name)
+            else:
+                raise ValueError("Duplicate test dataset name: {}".format(test_dataset_name))
+            test_set: BaseDataset = ComponentFactory.create(namespace.Stereotype.DATASET, test_dataset_name)
+            self.test_set_list.append(test_set)
+        self.print_info()
+        
+
+    def run(self):
+        Log.info("Loading from epoch {}.".format(self.current_epoch))
+        self.run_heuristic()
+        Log.success("Inference finished.")
+        
+
+    def run_heuristic(self):
+ 
+        test_set: BaseDataset
+        for dataset_idx, test_set in enumerate(self.test_set_list):
+            status_manager.set_progress("heuristic", "heuristic", f"dataset", dataset_idx, len(self.test_set_list))
+            test_set_name = test_set.get_name()
+
+            total=int(len(test_set))
+            for i in tqdm(range(total), desc=f"Processing {test_set_name}", ncols=100):
+                try:
+                    data = test_set.__getitem__(i)
+                    scene_name = data["scene_name"]
+                    inference_result_path = os.path.join(self.output_dir, test_set_name, f"{scene_name}.pkl")
+                    if os.path.exists(inference_result_path):
+                        Log.info(f"Inference result already exists for scene: {scene_name}")
+                        continue
+                    
+                    status_manager.set_progress("heuristic", "heuristic", f"Batch[{test_set_name}]", i+1, total)
+                    output = self.predict_sequence(data)
+                    self.save_inference_result(test_set_name, data["scene_name"], output)
+                except Exception as e:
+                    print(e)
+                    Log.error(f"Error, {e}")
+                    continue
+                
+        status_manager.set_progress("heuristic", "heuristic", f"dataset", len(self.test_set_list), len(self.test_set_list))
+        
+    def predict_sequence(self, data, cr_increase_threshold=0, overlap_area_threshold=25, scan_points_threshold=10, max_iter=5000, max_retry=5000, max_success=5000):
+        scene_name = data["scene_name"]
+        Log.info(f"Processing scene: {scene_name}")
+        status_manager.set_status("heuristic", "heuristic", "scene", scene_name)
+
+        ''' data for rendering '''
+        scene_path = data["scene_path"]
+        O_to_L_pose = data["O_to_L_pose"]
+        voxel_threshold = self.voxel_size
+        filter_degree = 75
+        down_sampled_model_pts = data["gt_pts"]
+        
+        first_frame_to_world_9d = data["first_scanned_n_to_world_pose_9d"][0]
+        first_frame_to_world = np.eye(4)
+        first_frame_to_world[:3,:3] = PoseUtil.rotation_6d_to_matrix_numpy(first_frame_to_world_9d[:6])
+        first_frame_to_world[:3,3] = first_frame_to_world_9d[6:]
+        
+        # 获取扫描点
+        root = os.path.dirname(scene_path)
+        display_table_info = DataLoadUtil.get_display_table_info(root, scene_name)
+        radius = display_table_info["radius"]
+        scan_points = np.asarray(ReconstructionUtil.generate_scan_points(display_table_top=0,display_table_radius=radius))
+
+        # 生成位姿序列
+        if self.heuristic_method == "hemisphere_random":
+            pose_sequence = self.generate_hemisphere_random_sequence(
+                max_iter,
+                self.heuristic_method_config
+            )
+        elif self.heuristic_method == "hemisphere_circle_trajectory":
+            pose_sequence = self.generate_hemisphere_circle_sequence(
+                self.heuristic_method_config
+            )
+        else:
+            raise ValueError(f"Unknown heuristic method: {self.heuristic_method}")
+
+        # 执行第一帧
+        first_frame_target_pts, _, first_frame_scan_points_indices = RenderUtil.render_pts(
+            first_frame_to_world, scene_path, self.script_path, scan_points, 
+            voxel_threshold=voxel_threshold, filter_degree=filter_degree, nO_to_nL_pose=O_to_L_pose
+        )
+        
+        # 初始化结果存储
+        scanned_view_pts = [first_frame_target_pts]
+        history_indices = [first_frame_scan_points_indices]
+        pred_cr_seq = []
+        retry_duplication_pose = []
+        retry_no_pts_pose = []
+        retry_overlap_pose = []
+        pose_9d_seq = [first_frame_to_world_9d]
+        
+        last_pred_cr, _ = self.compute_coverage_rate(scanned_view_pts, None, down_sampled_model_pts, threshold=voxel_threshold)
+        pred_cr_seq.append(last_pred_cr)
+        last_pts_num = PtsUtil.voxel_downsample_point_cloud(first_frame_target_pts, voxel_threshold).shape[0]
+        
+        # 执行序列
+        retry = 0
+        success = 0
+        #import ipdb; ipdb.set_trace()
+        combined_scanned_pts_tensor = torch.tensor([0,0,0])
+        cnt = 0
+        for pred_pose in pose_sequence:
+            cnt += 1
+            if retry >= max_retry or success >= max_success:
+                break
+            
+            Log.green(f"迭代: {cnt}/{len(pose_sequence)}, 重试: {retry}/{max_retry}, 成功: {success}/{max_success}")
+            
+            try:
+                new_target_pts, _, new_scan_points_indices = RenderUtil.render_pts(
+                    pred_pose, scene_path, self.script_path, scan_points,
+                    voxel_threshold=voxel_threshold, filter_degree=filter_degree, nO_to_nL_pose=O_to_L_pose
+                )
+                
+                # 检查扫描点重叠
+                if not ReconstructionUtil.check_scan_points_overlap(history_indices, new_scan_points_indices, scan_points_threshold):
+                    curr_overlap_area_threshold = overlap_area_threshold
+                else:
+                    curr_overlap_area_threshold = overlap_area_threshold * 0.5
+
+                # 检查点云重叠
+                downsampled_new_target_pts = PtsUtil.voxel_downsample_point_cloud(new_target_pts, voxel_threshold)
+                overlap, _ = ReconstructionUtil.check_overlap(
+                    downsampled_new_target_pts, down_sampled_model_pts,
+                    overlap_area_threshold=curr_overlap_area_threshold,
+                    voxel_size=voxel_threshold,
+                    require_new_added_pts_num=True
+                )
+                
+                if not overlap:
+                    Log.yellow("no overlap!")
+                    retry += 1
+                    retry_overlap_pose.append(pred_pose.tolist())
+                    continue
+
+                if new_target_pts.shape[0] == 0:
+                    Log.red("新视角无点云")
+                    retry_no_pts_pose.append(pred_pose.tolist())
+                    retry += 1
+                    continue
+                    
+                history_indices.append(new_scan_points_indices)
+                
+                # 计算覆盖率
+                pred_cr, _ = self.compute_coverage_rate(scanned_view_pts, new_target_pts, down_sampled_model_pts, threshold=voxel_threshold)
+                Log.yellow(f"覆盖率: {pred_cr}, 上一次: {last_pred_cr}, 最大: {data['seq_max_coverage_rate']}")
+                
+                # 更新结果
+                pred_cr_seq.append(pred_cr)
+                scanned_view_pts.append(new_target_pts)
+                pose_6d = PoseUtil.matrix_to_rotation_6d_numpy(pred_pose[:3,:3])
+                pose_9d = np.concatenate([
+                    pose_6d,
+                    pred_pose[:3,3]
+                ])
+                pose_9d_seq.append(pose_9d)
+                # 处理点云数据用于combined_scanned_pts
+                combined_scanned_pts = np.vstack(scanned_view_pts)
+                voxel_downsampled_pts, _ = self.voxel_downsample_with_mapping(combined_scanned_pts, voxel_threshold)
+                random_downsampled_pts, _ = PtsUtil.random_downsample_point_cloud(voxel_downsampled_pts, 8192, require_idx=True)
+                combined_scanned_pts_tensor = torch.tensor(random_downsampled_pts, dtype=torch.float32)
+
+                
+                # 检查点数增量
+                pts_num = voxel_downsampled_pts.shape[0]
+                Log.info(f"点数增量: {pts_num - last_pts_num}, 当前: {pts_num}, 上一次: {last_pts_num}")
+                
+                if pts_num - last_pts_num < self.min_new_pts_num:
+                    if pred_cr <= data["seq_max_coverage_rate"] - 1e-2:
+                        retry += 1
+                        retry_duplication_pose.append(pred_pose.tolist())
+                        Log.red(f"点数增量过小 < {self.min_new_pts_num}")
+                    else:
+                        success += 1
+                        Log.success(f"达到目标覆盖率")
+                        
+                last_pts_num = pts_num
+                last_pred_cr = pred_cr
+                
+                if pred_cr >= data["seq_max_coverage_rate"] - 1e-3:
+                    Log.success(f"达到最大覆盖率: {pred_cr}")
+                    
+                    
+            except Exception as e:
+                import traceback
+                traceback.print_exc()
+                Log.error(f"场景 {scene_path} 处理出错: {e}")
+                retry_no_pts_pose.append(pred_pose.tolist())
+                retry += 1
+                continue
+
+        # 返回结果
+        result = {
+            "pred_pose_9d_seq": pose_9d_seq,
+            "combined_scanned_pts_tensor": combined_scanned_pts_tensor,
+            "target_pts_seq": scanned_view_pts,
+            "coverage_rate_seq": pred_cr_seq,
+            "max_coverage_rate": data["seq_max_coverage_rate"],
+            "pred_max_coverage_rate": max(pred_cr_seq),
+            "scene_name": scene_name,
+            "retry_no_pts_pose": retry_no_pts_pose,
+            "retry_duplication_pose": retry_duplication_pose,
+            "retry_overlap_pose": retry_overlap_pose,
+            "best_seq_len": data["best_seq_len"],
+        }
+        
+        self.stat_result[scene_name] = {
+            "coverage_rate_seq": pred_cr_seq,
+            "pred_max_coverage_rate": max(pred_cr_seq),
+            "pred_seq_len": len(pred_cr_seq),
+        }
+        print('success rate: ', max(pred_cr_seq))
+
+        return result
+
+    def voxel_downsample_with_mapping(self, point_cloud, voxel_size=0.003):
+        voxel_indices = np.floor(point_cloud / voxel_size).astype(np.int32)
+        unique_voxels, inverse, counts = np.unique(voxel_indices, axis=0, return_inverse=True, return_counts=True)
+        idx_sort = np.argsort(inverse)
+        idx_unique = idx_sort[np.cumsum(counts)-counts]
+        downsampled_points = point_cloud[idx_unique]
+        return downsampled_points, inverse
+    
+    def compute_coverage_rate(self, scanned_view_pts, new_pts, model_pts, threshold=0.005):
+        if new_pts is not None:
+            new_scanned_view_pts = scanned_view_pts + [new_pts]
+        else:
+            new_scanned_view_pts = scanned_view_pts
+        combined_point_cloud = np.vstack(new_scanned_view_pts)
+        down_sampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_point_cloud,threshold)
+        return ReconstructionUtil.compute_coverage_rate(model_pts, down_sampled_combined_point_cloud, threshold)
+        
+
+    def save_inference_result(self, dataset_name, scene_name, output):
+        dataset_dir = os.path.join(self.output_dir, dataset_name)
+        if not os.path.exists(dataset_dir):
+            os.makedirs(dataset_dir)
+        output_path = os.path.join(dataset_dir, f"{scene_name}.pkl")
+        pickle.dump(output, open(output_path, "wb"))
+        with open(self.stat_result_path, "w") as f:
+            json.dump(self.stat_result, f)
+    
+
+    def get_checkpoint_path(self, is_last=False):
+        return os.path.join(self.experiment_path, namespace.Direcotry.CHECKPOINT_DIR_NAME,
+                            "Epoch_{}.pth".format(
+                                self.current_epoch if self.current_epoch != -1 and not is_last else "last"))
+
+    def load_checkpoint(self, is_last=False):
+        self.load(self.get_checkpoint_path(is_last))
+        Log.success(f"Loaded checkpoint from {self.get_checkpoint_path(is_last)}")
+        if is_last:
+            checkpoint_root = os.path.join(self.experiment_path, namespace.Direcotry.CHECKPOINT_DIR_NAME)
+            meta_path = os.path.join(checkpoint_root, "meta.json")
+            if not os.path.exists(meta_path):
+                raise FileNotFoundError(
+                    "No checkpoint meta.json file in the experiment {}".format(self.experiments_config["name"]))
+            file_path = os.path.join(checkpoint_root, "meta.json")
+            with open(file_path, "r") as f:
+                meta = json.load(f)
+            self.current_epoch = meta["last_epoch"]
+            self.current_iter = meta["last_iter"]
+
+    def load_experiment(self, backup_name=None):
+        super().load_experiment(backup_name)
+        self.current_epoch = self.experiments_config["epoch"]
+
+    def create_experiment(self, backup_name=None):
+        super().create_experiment(backup_name)
+       
+        
+    
+    def print_info(self):
+        def print_dataset(dataset: BaseDataset):
+            config = dataset.get_config()
+            name = dataset.get_name()
+            Log.blue(f"Dataset: {name}")
+            for k,v in config.items():
+                Log.blue(f"\t{k}: {v}")
+
+        super().print_info()
+        table_size = 70
+        Log.blue(f"{'+' + '-' * (table_size // 2)} Datasets {'-' * (table_size // 2)}" + '+')
+        for i, test_set in enumerate(self.test_set_list):
+            Log.blue(f"test dataset {i}: ")
+            print_dataset(test_set)
+
+        Log.blue(f"{'+' + '-' * (table_size // 2)}----------{'-' * (table_size // 2)}" + '+')
+
+    def generate_hemisphere_random_sequence(self, max_iter, config):
+        """Generate a random hemisphere sampling sequence"""
+        radius_fixed = config["radius_fixed"]
+        fixed_radius = config["fixed_radius"]
+        min_radius = config["min_radius"]
+        max_radius = config["max_radius"]
+        poses = []
+        center = np.array(config["center"])
+        
+        for _ in range(max_iter):
+            # 随机采样方向
+            direction = np.random.randn(3)
+            direction[2] = abs(direction[2])  # 确保在上半球
+            direction = direction / np.linalg.norm(direction)
+            
+            # 确定半径
+            if radius_fixed:
+                radius = fixed_radius
+            else:
+                radius = np.random.uniform(min_radius, max_radius)
+                
+            # 计算位置和朝向
+            position = center + direction * radius
+            z_axis = -direction
+            y_axis = np.array([0, 0, 1])
+            x_axis = np.cross(y_axis, z_axis)
+            x_axis = x_axis / np.linalg.norm(x_axis)
+            y_axis = np.cross(z_axis, x_axis)
+            
+            pose = np.eye(4)
+            pose[:3,:3] = np.stack([x_axis, y_axis, z_axis], axis=1)
+            pose[:3,3] = position
+            poses.append(pose)
+            
+        return poses
+
+    def generate_hemisphere_circle_sequence(self, config):
+        """Generate a circular trajectory sampling sequence"""
+        radius_fixed = config["radius_fixed"]
+        fixed_radius = config["fixed_radius"]
+        min_radius = config["min_radius"]
+        max_radius = config["max_radius"]
+        phi_list = config["phi_list"]
+        circle_times = config["circle_times"]
+        
+        poses = []
+        center = np.array(config["center"])
+        
+        for phi in phi_list:  # 仰角
+            phi_rad = np.deg2rad(phi)
+            for i in range(circle_times):  # 方位角
+                theta = i * (2 * np.pi / circle_times)
+                
+                # 确定半径
+                if radius_fixed:
+                    radius = fixed_radius
+                else:
+                    radius = np.random.uniform(min_radius, max_radius)
+                    
+                # 球坐标转笛卡尔坐标
+                x = radius * np.cos(theta) * np.sin(phi_rad)
+                y = radius * np.sin(theta) * np.sin(phi_rad)
+                z = radius * np.cos(phi_rad)
+                position = center + np.array([x, y, z])
+                
+                # 计算朝向
+                direction = (center - position) / np.linalg.norm(center - position)
+                z_axis = direction
+                y_axis = np.array([0, 0, 1])
+                x_axis = np.cross(y_axis, z_axis)
+                x_axis = x_axis / np.linalg.norm(x_axis)
+                y_axis = np.cross(z_axis, x_axis)
+                
+                pose = np.eye(4)
+                pose[:3,:3] = np.stack([x_axis, y_axis, z_axis], axis=1)
+                pose[:3,3] = position
+                poses.append(pose)
+                
+        return poses
+

runners/inference_server.py

@@ -0,0 +1,114 @@
+import os
+import json
+import torch
+import numpy as np
+from flask import Flask, request, jsonify
+
+import PytorchBoot.namespace as namespace
+import PytorchBoot.stereotype as stereotype
+from PytorchBoot.factory import ComponentFactory
+
+from PytorchBoot.runners.runner import Runner
+from PytorchBoot.utils import Log
+
+from utils.pts import PtsUtil
+
+@stereotype.runner("inferencer_server")
+class InferencerServer(Runner):
+    def __init__(self, config_path):
+        super().__init__(config_path)
+    
+        ''' Web Server '''
+        self.app = Flask(__name__)
+        ''' Pipeline '''
+        self.pipeline_name = self.config[namespace.Stereotype.PIPELINE]
+        self.pipeline:torch.nn.Module = ComponentFactory.create(namespace.Stereotype.PIPELINE, self.pipeline_name)
+        self.pipeline = self.pipeline.to(self.device)
+        self.pts_num = 8192
+        self.voxel_size = 0.002
+            
+        ''' Experiment '''
+        self.load_experiment("inferencer_server")
+
+    def get_input_data(self, data):
+        input_data = {}
+        scanned_pts = data["scanned_pts"]
+        scanned_n_to_world_pose_9d = data["scanned_n_to_world_pose_9d"]
+        combined_scanned_views_pts = np.concatenate(scanned_pts, axis=0)
+        voxel_downsampled_combined_scanned_pts = PtsUtil.voxel_downsample_point_cloud(
+            combined_scanned_views_pts, self.voxel_size
+        )
+        fps_downsampled_combined_scanned_pts, fps_idx = PtsUtil.fps_downsample_point_cloud(
+            voxel_downsampled_combined_scanned_pts, self.pts_num, require_idx=True
+        )
+
+        input_data["scanned_pts"] = scanned_pts
+        input_data["scanned_n_to_world_pose_9d"] = np.asarray(scanned_n_to_world_pose_9d, dtype=np.float32)
+        input_data["combined_scanned_pts"] = np.asarray(fps_downsampled_combined_scanned_pts, dtype=np.float32)
+        return input_data
+    
+    def get_result(self, output_data):
+
+        pred_pose_9d = output_data["pred_pose_9d"]
+        result = {
+            "pred_pose_9d": pred_pose_9d.tolist()
+        }
+        return result
+    
+    def collate_input(self, input_data):
+        collated_input_data = {}
+        collated_input_data["scanned_pts"] = [torch.tensor(input_data["scanned_pts"], dtype=torch.float32, device=self.device)]
+        collated_input_data["scanned_n_to_world_pose_9d"] = [torch.tensor(input_data["scanned_n_to_world_pose_9d"], dtype=torch.float32, device=self.device)]
+        collated_input_data["combined_scanned_pts"] = torch.tensor(input_data["combined_scanned_pts"], dtype=torch.float32, device=self.device).unsqueeze(0)
+        return collated_input_data
+    
+    def run(self):
+        Log.info("Loading from epoch {}.".format(self.current_epoch))
+    
+        @self.app.route("/inference", methods=["POST"])
+        def inference():
+            data = request.json
+            input_data = self.get_input_data(data)
+            collated_input_data = self.collate_input(input_data)
+            output_data = self.pipeline.forward_test(collated_input_data)
+            result = self.get_result(output_data)
+            return jsonify(result)
+        
+
+        self.app.run(host="0.0.0.0", port=5000)
+
+    def get_checkpoint_path(self, is_last=False):
+        return os.path.join(self.experiment_path, namespace.Direcotry.CHECKPOINT_DIR_NAME,
+                            "Epoch_{}.pth".format(
+                                self.current_epoch if self.current_epoch != -1 and not is_last else "last"))
+
+    def load_checkpoint(self, is_last=False):
+        self.load(self.get_checkpoint_path(is_last))
+        Log.success(f"Loaded checkpoint from {self.get_checkpoint_path(is_last)}")
+        if is_last:
+            checkpoint_root = os.path.join(self.experiment_path, namespace.Direcotry.CHECKPOINT_DIR_NAME)
+            meta_path = os.path.join(checkpoint_root, "meta.json")
+            if not os.path.exists(meta_path):
+                raise FileNotFoundError(
+                    "No checkpoint meta.json file in the experiment {}".format(self.experiments_config["name"]))
+            file_path = os.path.join(checkpoint_root, "meta.json")
+            with open(file_path, "r") as f:
+                meta = json.load(f)
+            self.current_epoch = meta["last_epoch"]
+            self.current_iter = meta["last_iter"]
+
+    def load_experiment(self, backup_name=None):
+        super().load_experiment(backup_name)
+        self.current_epoch = self.experiments_config["epoch"]
+        self.load_checkpoint(is_last=(self.current_epoch == -1))
+
+    def create_experiment(self, backup_name=None):
+        super().create_experiment(backup_name)
+       
+        
+    def load(self, path):
+        state_dict = torch.load(path)
+        self.pipeline.load_state_dict(state_dict)
+
+
+

runners/inferencer.py

@@ -19,14 +19,19 @@ from PytorchBoot.dataset import BaseDataset
 from PytorchBoot.runners.runner import Runner
 from PytorchBoot.utils import Log
 from PytorchBoot.status import status_manager
-
+from utils.data_load import DataLoadUtil
 @stereotype.runner("inferencer")
 class Inferencer(Runner):
     def __init__(self, config_path):
+        
         super().__init__(config_path)
     
         self.script_path = ConfigManager.get(namespace.Stereotype.RUNNER, "blender_script_path")
         self.output_dir = ConfigManager.get(namespace.Stereotype.RUNNER, "output_dir")
+        self.voxel_size = ConfigManager.get(namespace.Stereotype.RUNNER, "voxel_size")
+        self.min_new_area = ConfigManager.get(namespace.Stereotype.RUNNER, "min_new_area")
+        CM = 0.01
+        self.min_new_pts_num = self.min_new_area * (CM / self.voxel_size) **2
         ''' Pipeline '''
         self.pipeline_name = self.config[namespace.Stereotype.PIPELINE]
         self.pipeline:torch.nn.Module = ComponentFactory.create(namespace.Stereotype.PIPELINE, self.pipeline_name)
@@ -34,7 +39,12 @@ class Inferencer(Runner):
             
         ''' Experiment '''
         self.load_experiment("nbv_evaluator")
-        self.stat_result = {}
+        self.stat_result_path = os.path.join(self.output_dir, "stat.json")
+        if os.path.exists(self.stat_result_path):
+            with open(self.stat_result_path, "r") as f:
+                self.stat_result = json.load(f)
+        else:
+            self.stat_result = {}
 
         ''' Test '''
         self.test_config = ConfigManager.get(namespace.Stereotype.RUNNER, namespace.Mode.TEST)
@@ -65,58 +75,74 @@ class Inferencer(Runner):
             for dataset_idx, test_set in enumerate(self.test_set_list):
                 status_manager.set_progress("inference", "inferencer", f"dataset", dataset_idx, len(self.test_set_list))
                 test_set_name = test_set.get_name()
-                test_loader = test_set.get_loader()
 
-                if test_loader.batch_size > 1:
-                    Log.error("Batch size should be 1 for inference, found {} in {}".format(test_loader.batch_size, test_set_name), terminate=True)
-
-                total=int(len(test_loader))
-                loop = tqdm(enumerate(test_loader), total=total)
-                for i, data in loop:
-                    status_manager.set_progress("inference", "inferencer", f"Batch[{test_set_name}]", i+1, total)
-                    test_set.process_batch(data, self.device)
-                    output = self.predict_sequence(data)
-                    self.save_inference_result(test_set_name, data["scene_name"][0], output)
+                total=int(len(test_set))
+                for i in tqdm(range(total), desc=f"Processing {test_set_name}", ncols=100):
+                    try:
+                        data = test_set.__getitem__(i)
+                        scene_name = data["scene_name"]
+                        inference_result_path = os.path.join(self.output_dir, test_set_name, f"{scene_name}.pkl")
+                        if os.path.exists(inference_result_path):
+                            Log.info(f"Inference result already exists for scene: {scene_name}")
+                            continue
+                        
+                        status_manager.set_progress("inference", "inferencer", f"Batch[{test_set_name}]", i+1, total)
+                        output = self.predict_sequence(data)
+                        self.save_inference_result(test_set_name, data["scene_name"], output)
+                    except Exception as e:
+                        print(e)
+                        Log.error(f"Error, {e}")
+                        continue
                     
             status_manager.set_progress("inference", "inferencer", f"dataset", len(self.test_set_list), len(self.test_set_list))
         
-    def predict_sequence(self, data, cr_increase_threshold=0, max_iter=50, max_retry=5):
-        scene_name = data["scene_name"][0]
+    def predict_sequence(self, data, cr_increase_threshold=0, overlap_area_threshold=25, scan_points_threshold=10, max_iter=50, max_retry = 10, max_success=3):
+        scene_name = data["scene_name"]
         Log.info(f"Processing scene: {scene_name}")
         status_manager.set_status("inference", "inferencer", "scene", scene_name)
 
         ''' data for rendering '''
-        scene_path = data["scene_path"][0]
-        O_to_L_pose = data["O_to_L_pose"][0]
-        voxel_threshold = data["voxel_threshold"][0]
-        filter_degree = data["filter_degree"][0]
-        model_points_normals = data["model_points_normals"][0]
-        model_pts = model_points_normals[:,:3]
-        down_sampled_model_pts = PtsUtil.voxel_downsample_point_cloud(model_pts, voxel_threshold)
-        first_frame_to_world_9d = data["first_to_world_9d"][0]
-        first_frame_to_world = torch.eye(4, device=first_frame_to_world_9d.device)
-        first_frame_to_world[:3,:3] = PoseUtil.rotation_6d_to_matrix_tensor_batch(first_frame_to_world_9d[:,:6])[0]
-        first_frame_to_world[:3,3] = first_frame_to_world_9d[0,6:]
-        first_frame_to_world = first_frame_to_world.to(self.device)
+        scene_path = data["scene_path"]
+        O_to_L_pose = data["O_to_L_pose"]
+        voxel_threshold = self.voxel_size
+        filter_degree = 75
+        down_sampled_model_pts = data["gt_pts"]
+        
+        first_frame_to_world_9d = data["first_scanned_n_to_world_pose_9d"][0]
+        first_frame_to_world = np.eye(4)
+        first_frame_to_world[:3,:3] = PoseUtil.rotation_6d_to_matrix_numpy(first_frame_to_world_9d[:6])
+        first_frame_to_world[:3,3] = first_frame_to_world_9d[6:]
         
         ''' data for inference '''
         input_data = {}
-        input_data["scanned_pts"] = [data["first_pts"][0].to(self.device)]
-        input_data["scanned_n_to_world_pose_9d"] = [data["first_to_world_9d"][0].to(self.device)]
-        input_data["mode"] = namespace.Mode.TEST
-        input_data["combined_scanned_pts"] = data["combined_scanned_pts"]
-        input_pts_N = input_data["scanned_pts"][0].shape[1]
         
-        first_frame_target_pts, _ = RenderUtil.render_pts(first_frame_to_world, scene_path, self.script_path, model_points_normals, voxel_threshold=voxel_threshold, filter_degree=filter_degree, nO_to_nL_pose=O_to_L_pose)
-        scanned_view_pts = [first_frame_target_pts]
-        last_pred_cr = self.compute_coverage_rate(scanned_view_pts, None, down_sampled_model_pts, threshold=voxel_threshold)
+        input_data["combined_scanned_pts"] = torch.tensor(data["first_scanned_pts"][0], dtype=torch.float32).to(self.device).unsqueeze(0)
+        input_data["scanned_pts_mask"] = [torch.zeros(input_data["combined_scanned_pts"].shape[1], dtype=torch.bool).to(self.device).unsqueeze(0)]
+        input_data["scanned_n_to_world_pose_9d"] = [torch.tensor(data["first_scanned_n_to_world_pose_9d"], dtype=torch.float32).to(self.device)]
+        input_data["mode"] = namespace.Mode.TEST
+        input_pts_N = input_data["combined_scanned_pts"].shape[1]
+        
+        root = os.path.dirname(scene_path)
+        display_table_info = DataLoadUtil.get_display_table_info(root, scene_name)
+        radius = display_table_info["radius"]
+        scan_points = np.asarray(ReconstructionUtil.generate_scan_points(display_table_top=0,display_table_radius=radius))
 
+        first_frame_target_pts, first_frame_target_normals, first_frame_scan_points_indices = RenderUtil.render_pts(first_frame_to_world, scene_path, self.script_path, scan_points, voxel_threshold=voxel_threshold, filter_degree=filter_degree, nO_to_nL_pose=O_to_L_pose)
+        scanned_view_pts = [first_frame_target_pts]
+        history_indices = [first_frame_scan_points_indices]
+        last_pred_cr, added_pts_num = self.compute_coverage_rate(scanned_view_pts, None, down_sampled_model_pts, threshold=voxel_threshold)
         retry_duplication_pose = []
         retry_no_pts_pose = []
+        retry_overlap_pose = []
         retry = 0
         pred_cr_seq = [last_pred_cr]
-        while len(pred_cr_seq) < max_iter and retry < max_retry:
-            
+        success = 0
+        last_pts_num = PtsUtil.voxel_downsample_point_cloud(data["first_scanned_pts"][0], voxel_threshold).shape[0]
+        import time
+        while len(pred_cr_seq) < max_iter and retry < max_retry and success < max_success:
+            Log.green(f"iter: {len(pred_cr_seq)}, retry: {retry}/{max_retry}, success: {success}/{max_success}")
+            combined_scanned_pts = np.vstack(scanned_view_pts)
+            voxel_downsampled_combined_scanned_pts_np, inverse =  self.voxel_downsample_with_mapping(combined_scanned_pts, voxel_threshold)
             output = self.pipeline(input_data)
             pred_pose_9d = output["pred_pose_9d"]
             pred_pose = torch.eye(4, device=pred_pose_9d.device)
@@ -125,71 +151,119 @@ class Inferencer(Runner):
             pred_pose[:3,3] = pred_pose_9d[0,6:]
             
             try:
-                new_target_pts_world, new_pts_world = RenderUtil.render_pts(pred_pose, scene_path, self.script_path, model_points_normals, voxel_threshold=voxel_threshold, filter_degree=filter_degree, nO_to_nL_pose=O_to_L_pose, require_full_scene=True)
+                new_target_pts, new_target_normals, new_scan_points_indices = RenderUtil.render_pts(pred_pose, scene_path, self.script_path, scan_points, voxel_threshold=voxel_threshold, filter_degree=filter_degree, nO_to_nL_pose=O_to_L_pose)
+                #import ipdb; ipdb.set_trace()
+                if not ReconstructionUtil.check_scan_points_overlap(history_indices, new_scan_points_indices, scan_points_threshold):
+                    curr_overlap_area_threshold = overlap_area_threshold
+                else:
+                    curr_overlap_area_threshold = overlap_area_threshold * 0.5  
+
+                downsampled_new_target_pts = PtsUtil.voxel_downsample_point_cloud(new_target_pts, voxel_threshold)
+                overlap, _ = ReconstructionUtil.check_overlap(downsampled_new_target_pts, voxel_downsampled_combined_scanned_pts_np, overlap_area_threshold = curr_overlap_area_threshold, voxel_size=voxel_threshold, require_new_added_pts_num = True)
+                if not overlap:
+                    Log.yellow("no overlap!")
+                    retry += 1
+                    retry_overlap_pose.append(pred_pose.cpu().numpy().tolist())
+                    continue
+                
+                history_indices.append(new_scan_points_indices)
             except Exception as e:
-                Log.warning(f"Error in scene {scene_path}, {e}")
+                Log.error(f"Error in scene {scene_path}, {e}")
                 print("current pose: ", pred_pose)
                 print("curr_pred_cr: ", last_pred_cr)
                 retry_no_pts_pose.append(pred_pose.cpu().numpy().tolist())
                 retry += 1
                 continue
             
-            
-            pred_cr = self.compute_coverage_rate(scanned_view_pts, new_target_pts_world, down_sampled_model_pts, threshold=voxel_threshold)
-            
-            print(pred_cr, last_pred_cr, " max: ", data["max_coverage_rate"])
-            if pred_cr >= data["max_coverage_rate"]:
-                print("max coverage rate reached!")
-            if pred_cr <= last_pred_cr + cr_increase_threshold:
+            if new_target_pts.shape[0] == 0:
+                Log.red("no pts in new target")
+                retry_no_pts_pose.append(pred_pose.cpu().numpy().tolist())
                 retry += 1
-                retry_duplication_pose.append(pred_pose.cpu().numpy().tolist())
                 continue
-
-            retry = 0
-            pred_cr_seq.append(pred_cr)
-            scanned_view_pts.append(new_target_pts_world)
-            down_sampled_new_pts_world = PtsUtil.random_downsample_point_cloud(new_pts_world, input_pts_N)
-            new_pts_world_aug = np.hstack([down_sampled_new_pts_world, np.ones((down_sampled_new_pts_world.shape[0], 1))])
-            new_pts = np.dot(np.linalg.inv(first_frame_to_world.cpu()), new_pts_world_aug.T).T[:,:3]
-
-            new_pts_tensor = torch.tensor(new_pts, dtype=torch.float32).unsqueeze(0).to(self.device)
             
-            input_data["scanned_pts"] = [torch.cat([input_data["scanned_pts"][0] , new_pts_tensor], dim=0)]
+            pred_cr, _ = self.compute_coverage_rate(scanned_view_pts, new_target_pts, down_sampled_model_pts, threshold=voxel_threshold)
+            Log.yellow(f"{pred_cr}, {last_pred_cr}, max: , {data['seq_max_coverage_rate']}")
+            if pred_cr >= data["seq_max_coverage_rate"] - 1e-3:
+                print("max coverage rate reached!: ", pred_cr)
+                
+        
+
+            pred_cr_seq.append(pred_cr)
+            scanned_view_pts.append(new_target_pts)
+            
             input_data["scanned_n_to_world_pose_9d"] = [torch.cat([input_data["scanned_n_to_world_pose_9d"][0], pred_pose_9d], dim=0)]
-            combined_scanned_views_pts = np.concatenate(input_data["scanned_pts"][0].tolist(), axis=0)
-            voxel_downsampled_combined_scanned_pts_np = PtsUtil.voxel_downsample_point_cloud(combined_scanned_views_pts, 0.002)
-            random_downsampled_combined_scanned_pts_np = PtsUtil.random_downsample_point_cloud(voxel_downsampled_combined_scanned_pts_np, input_pts_N)
+            start_indices = [0]
+            total_points = 0
+            for pts in scanned_view_pts:
+                total_points += pts.shape[0]
+                start_indices.append(total_points)
+            combined_scanned_pts = np.vstack(scanned_view_pts)
+            voxel_downsampled_combined_scanned_pts_np, inverse =  self.voxel_downsample_with_mapping(combined_scanned_pts, voxel_threshold)
+            random_downsampled_combined_scanned_pts_np, random_downsample_idx = PtsUtil.random_downsample_point_cloud(voxel_downsampled_combined_scanned_pts_np, input_pts_N, require_idx=True)
+            all_idx_unique = np.arange(len(voxel_downsampled_combined_scanned_pts_np))
+            all_random_downsample_idx = all_idx_unique[random_downsample_idx]
+            scanned_pts_mask = []
+            for idx, start_idx in enumerate(start_indices):
+                if idx == len(start_indices) - 1:
+                    break
+                end_idx = start_indices[idx+1]
+                view_inverse = inverse[start_idx:end_idx]
+                view_unique_downsampled_idx = np.unique(view_inverse)
+                view_unique_downsampled_idx_set = set(view_unique_downsampled_idx)
+                mask = np.array([idx in view_unique_downsampled_idx_set for idx in all_random_downsample_idx])
+                scanned_pts_mask.append(mask)
+
             input_data["combined_scanned_pts"] = torch.tensor(random_downsampled_combined_scanned_pts_np, dtype=torch.float32).unsqueeze(0).to(self.device)
+            #import ipdb; ipdb.set_trace()
+            input_data["scanned_pts_mask"] = [torch.tensor(scanned_pts_mask, dtype=torch.bool)]
 
             
             last_pred_cr = pred_cr
+            pts_num = voxel_downsampled_combined_scanned_pts_np.shape[0]
+            Log.info(f"delta pts num:,{pts_num - last_pts_num },{pts_num}, {last_pts_num}")
+
+            if pts_num - last_pts_num < self.min_new_pts_num and pred_cr <= data["seq_max_coverage_rate"] - 1e-2:
+                retry += 1
+                retry_duplication_pose.append(pred_pose.cpu().numpy().tolist())
+                Log.red(f"delta pts num < {self.min_new_pts_num}:, {pts_num}, {last_pts_num}")
+            elif pts_num - last_pts_num < self.min_new_pts_num and pred_cr > data["seq_max_coverage_rate"] - 1e-2:
+                success += 1
+                Log.success(f"delta pts num < {self.min_new_pts_num}:, {pts_num}, {last_pts_num}")
+
+            last_pts_num = pts_num
+            
 
-        
-        input_data["scanned_pts"] = input_data["scanned_pts"][0].cpu().numpy().tolist()
         input_data["scanned_n_to_world_pose_9d"] = input_data["scanned_n_to_world_pose_9d"][0].cpu().numpy().tolist()
         result = {
             "pred_pose_9d_seq": input_data["scanned_n_to_world_pose_9d"],
-            "pts_seq": input_data["scanned_pts"],
+            "combined_scanned_pts": input_data["combined_scanned_pts"],
             "target_pts_seq": scanned_view_pts,
             "coverage_rate_seq": pred_cr_seq,
-            "max_coverage_rate": data["max_coverage_rate"][0],
+            "max_coverage_rate": data["seq_max_coverage_rate"],
             "pred_max_coverage_rate": max(pred_cr_seq),
             "scene_name": scene_name,
             "retry_no_pts_pose": retry_no_pts_pose,
             "retry_duplication_pose": retry_duplication_pose,
-            "best_seq_len": data["best_seq_len"][0],
+            "retry_overlap_pose": retry_overlap_pose,
+            "best_seq_len": data["best_seq_len"],
         }
         self.stat_result[scene_name] = {
-            "max_coverage_rate": data["max_coverage_rate"][0],
-            "success_rate": max(pred_cr_seq)/ data["max_coverage_rate"][0],
             "coverage_rate_seq": pred_cr_seq,
             "pred_max_coverage_rate": max(pred_cr_seq),
             "pred_seq_len": len(pred_cr_seq),
         }
-        print('success rate: ', max(pred_cr_seq) / data["max_coverage_rate"][0])
+        print('success rate: ', max(pred_cr_seq))
 
         return result
 
+    def voxel_downsample_with_mapping(self, point_cloud, voxel_size=0.003):
+        voxel_indices = np.floor(point_cloud / voxel_size).astype(np.int32)
+        unique_voxels, inverse, counts = np.unique(voxel_indices, axis=0, return_inverse=True, return_counts=True)
+        idx_sort = np.argsort(inverse)
+        idx_unique = idx_sort[np.cumsum(counts)-counts]
+        downsampled_points = point_cloud[idx_unique]
+        return downsampled_points, inverse
+    
     def compute_coverage_rate(self, scanned_view_pts, new_pts, model_pts, threshold=0.005):
         if new_pts is not None:
             new_scanned_view_pts = scanned_view_pts + [new_pts]
@@ -206,7 +280,7 @@ class Inferencer(Runner):
             os.makedirs(dataset_dir)
         output_path = os.path.join(dataset_dir, f"{scene_name}.pkl")
         pickle.dump(output, open(output_path, "wb"))
-        with open(os.path.join(dataset_dir, "stat.json"), "w") as f:
+        with open(self.stat_result_path, "w") as f:
             json.dump(self.stat_result, f)
     
 

runners/strategy_generator.py

@@ -22,25 +22,21 @@ class StrategyGenerator(Runner):
             "app_name": "generate_strategy",
             "runner_name": "strategy_generator"
         }
-        self.to_specified_dir = ConfigManager.get("runner", "generate", "to_specified_dir")
-        self.save_best_combined_pts = ConfigManager.get("runner", "generate", "save_best_combined_points")
-        self.save_mesh = ConfigManager.get("runner", "generate", "save_mesh")
-        self.load_pts = ConfigManager.get("runner", "generate", "load_points")
-        self.filter_degree = ConfigManager.get("runner", "generate", "filter_degree")
         self.overwrite = ConfigManager.get("runner", "generate", "overwrite")
-        self.save_pts = ConfigManager.get("runner","generate","save_points")
         self.seq_num = ConfigManager.get("runner","generate","seq_num")
+        self.overlap_area_threshold = ConfigManager.get("runner","generate","overlap_area_threshold")
+        self.compute_with_normal = ConfigManager.get("runner","generate","compute_with_normal")
+        self.scan_points_threshold = ConfigManager.get("runner","generate","scan_points_threshold")
         
 
         
     def run(self):
         dataset_name_list =  ConfigManager.get("runner", "generate", "dataset_list")
-        voxel_threshold, soft_overlap_threshold, hard_overlap_threshold = ConfigManager.get("runner","generate","voxel_threshold"), ConfigManager.get("runner","generate","soft_overlap_threshold"), ConfigManager.get("runner","generate","hard_overlap_threshold")
+        voxel_threshold = ConfigManager.get("runner","generate","voxel_threshold")
         for dataset_idx in range(len(dataset_name_list)):
             dataset_name = dataset_name_list[dataset_idx]
             status_manager.set_progress("generate_strategy", "strategy_generator", "dataset", dataset_idx, len(dataset_name_list))
             root_dir = ConfigManager.get("datasets", dataset_name, "root_dir")
-            model_dir = ConfigManager.get("datasets", dataset_name, "model_dir")
             from_idx = ConfigManager.get("datasets",dataset_name,"from")
             to_idx = ConfigManager.get("datasets",dataset_name,"to")
             scene_name_list = os.listdir(root_dir)
@@ -52,17 +48,13 @@ class StrategyGenerator(Runner):
             for scene_name in scene_name_list[from_idx:to_idx]:
                 Log.info(f"({dataset_name})Processing [{cnt}/{total}]: {scene_name}")
                 status_manager.set_progress("generate_strategy", "strategy_generator", "scene", cnt, total)
-                diag = DataLoadUtil.get_bbox_diag(model_dir, scene_name)
-                status_manager.set_status("generate_strategy", "strategy_generator", "diagonal", diag)
                 output_label_path = DataLoadUtil.get_label_path(root_dir, scene_name,0)
                 if os.path.exists(output_label_path) and not self.overwrite:
                     Log.info(f"Scene <{scene_name}> Already Exists, Skip")
                     cnt += 1
                     continue
                 
-                self.generate_sequence(root_dir, model_dir, scene_name,voxel_threshold, soft_overlap_threshold, hard_overlap_threshold)
-                # except Exception as e:
-                #     Log.error(f"Scene <{scene_name}> Failed, Error: {e}")
+                self.generate_sequence(root_dir, scene_name,voxel_threshold)
                 cnt += 1
             status_manager.set_progress("generate_strategy", "strategy_generator", "scene", total, total)
         status_manager.set_progress("generate_strategy", "strategy_generator", "dataset", len(dataset_name_list), len(dataset_name_list))
@@ -75,36 +67,61 @@ class StrategyGenerator(Runner):
     def load_experiment(self, backup_name=None):
         super().load_experiment(backup_name)
     
-    def generate_sequence(self, root, model_dir, scene_name, voxel_threshold, soft_overlap_threshold, hard_overlap_threshold):
+    def generate_sequence(self, root, scene_name, voxel_threshold):
         status_manager.set_status("generate_strategy", "strategy_generator", "scene", scene_name)
         frame_num = DataLoadUtil.get_scene_seq_length(root, scene_name)
+        
         model_points_normals = DataLoadUtil.load_points_normals(root, scene_name)
         model_pts = model_points_normals[:,:3]
-        down_sampled_model_pts = PtsUtil.voxel_downsample_point_cloud(model_pts, voxel_threshold)
+        down_sampled_model_pts, idx = PtsUtil.voxel_downsample_point_cloud(model_pts, voxel_threshold, require_idx=True)
+        down_sampled_model_nrm = model_points_normals[idx, 3:]
         pts_list = []
+        nrm_list = []
         scan_points_indices_list = []
         non_zero_cnt = 0
+
         for frame_idx in range(frame_num):
             status_manager.set_progress("generate_strategy", "strategy_generator", "loading frame", frame_idx, frame_num)
-            pts_path = os.path.join(root,scene_name, "target_pts", f"{frame_idx}.txt")
-            sampled_point_cloud = np.loadtxt(pts_path)
-            indices = None # ReconstructionUtil.compute_covered_scan_points(scan_points, display_table_pts)
-            pts_list.append(sampled_point_cloud)
+            pts_path = os.path.join(root,scene_name, "pts", f"{frame_idx}.npy")
+            nrm_path = os.path.join(root,scene_name, "nrm", f"{frame_idx}.npy")
+            idx_path = os.path.join(root,scene_name, "scan_points_indices", f"{frame_idx}.npy")
+            
+            pts = np.load(pts_path)
+            if self.compute_with_normal:
+                if pts.shape[0] == 0:
+                    nrm = np.zeros((0,3))
+                else:
+                    nrm = np.load(nrm_path)
+                nrm_list.append(nrm)
+            pts_list.append(pts)
+            indices = np.load(idx_path) 
             scan_points_indices_list.append(indices)
+            if pts.shape[0] > 0:
+                non_zero_cnt += 1
         status_manager.set_progress("generate_strategy", "strategy_generator", "loading frame", frame_num, frame_num)
-
+        
         seq_num = min(self.seq_num, non_zero_cnt)
         init_view_list = []
-        for i in range(seq_num):
-            if pts_list[i].shape[0] < 100:
-                continue
-            init_view_list.append(i)
+        idx = 0
+        while len(init_view_list) < seq_num and idx < len(pts_list):
+            if pts_list[idx].shape[0] > 50:
+                init_view_list.append(idx)
+            idx += 1
 
         seq_idx = 0
+        import time
         for init_view in init_view_list:
             status_manager.set_progress("generate_strategy", "strategy_generator", "computing sequence", seq_idx, len(init_view_list))
-            limited_useful_view, _, _ = ReconstructionUtil.compute_next_best_view_sequence_with_overlap(down_sampled_model_pts, pts_list, scan_points_indices_list = scan_points_indices_list,init_view=init_view, 
-                                                                                                        threshold=voxel_threshold, soft_overlap_threshold=soft_overlap_threshold, hard_overlap_threshold= hard_overlap_threshold, scan_points_threshold=10, status_info=self.status_info)
+            start = time.time()
+
+            if not self.compute_with_normal:    
+                limited_useful_view, _, _ = ReconstructionUtil.compute_next_best_view_sequence(down_sampled_model_pts, pts_list, scan_points_indices_list = scan_points_indices_list,init_view=init_view, 
+                                                                                                            threshold=voxel_threshold, scan_points_threshold=self.scan_points_threshold, overlap_area_threshold=self.overlap_area_threshold, status_info=self.status_info)
+            else:
+                limited_useful_view, _, _ = ReconstructionUtil.compute_next_best_view_sequence_with_normal(down_sampled_model_pts, down_sampled_model_nrm, pts_list, nrm_list, scan_points_indices_list = scan_points_indices_list,init_view=init_view, 
+                                                                                                            threshold=voxel_threshold, scan_points_threshold=self.scan_points_threshold, overlap_area_threshold=self.overlap_area_threshold, status_info=self.status_info)
+            end = time.time()
+            print(f"Time: {end-start}")
             data_pairs = self.generate_data_pairs(limited_useful_view)
             seq_save_data = {
                 "data_pairs": data_pairs,
@@ -121,8 +138,6 @@ class StrategyGenerator(Runner):
             with open(output_label_path, 'w') as f:
                 json.dump(seq_save_data, f)
             seq_idx += 1
-        if self.save_mesh:
-            DataLoadUtil.save_target_mesh_at_world_space(root, model_dir, scene_name)
         status_manager.set_progress("generate_strategy", "strategy_generator", "computing sequence", len(init_view_list), len(init_view_list))
 
 

runners/view_generator.py

@@ -9,7 +9,8 @@ class ViewGenerator(Runner):
         self.config_path = config_path
         
     def run(self):
-        subprocess.run(['blender', '-b', '-P', '../blender/run_blender.py', '--', self.config_path])
+        result = subprocess.run(['blender', '-b', '-P', '../blender/run_blender.py', '--', self.config_path])
+        print()
 
     def create_experiment(self, backup_name=None):
         return super().create_experiment(backup_name)

utils/data_load.py

@@ -4,11 +4,28 @@ import json
 import cv2
 import trimesh
 import torch
+import OpenEXR
+import Imath
 from utils.pts import PtsUtil
 
 
 class DataLoadUtil:
     TABLE_POSITION = np.asarray([0, 0, 0.8215])
+    
+    @staticmethod
+    def load_exr_image(file_path):
+        exr_file = OpenEXR.InputFile(file_path)
+        header = exr_file.header()
+        dw = header['dataWindow']
+        width = dw.max.x - dw.min.x + 1
+        height = dw.max.y - dw.min.y + 1
+        float_channels = ['R', 'G', 'B']
+        img_data = []
+        for channel in float_channels:
+            channel_data = exr_file.channel(channel)
+            img_data.append(np.frombuffer(channel_data, dtype=np.float16).reshape((height, width)))
+        img = np.stack(img_data, axis=-1)
+        return img
 
     @staticmethod
     def get_display_table_info(root, scene_name):
@@ -34,6 +51,8 @@ class DataLoadUtil:
     @staticmethod
     def get_label_num(root, scene_name):
         label_dir = os.path.join(root, scene_name, "label")
+        if not os.path.exists(label_dir):
+            return 0
         return len(os.listdir(label_dir))
 
     @staticmethod
@@ -44,11 +63,6 @@ class DataLoadUtil:
         path = os.path.join(label_dir, f"{seq_idx}.json")
         return path
 
-    @staticmethod
-    def get_label_path_old(root, scene_name):
-        path = os.path.join(root, scene_name, "label.json")
-        return path
-
     @staticmethod
     def get_scene_seq_length(root, scene_name):
         camera_params_path = os.path.join(root, scene_name, "camera_params")
@@ -69,36 +83,6 @@ class DataLoadUtil:
         diagonal_length = np.linalg.norm(bbox)
         return diagonal_length
 
-    @staticmethod
-    def save_mesh_at(model_dir, output_dir, object_name, scene_name, world_object_pose):
-        mesh = DataLoadUtil.load_mesh_at(model_dir, object_name, world_object_pose)
-        model_path = os.path.join(output_dir, scene_name, "world_mesh.obj")
-        mesh.export(model_path)
-
-    @staticmethod
-    def save_target_mesh_at_world_space(
-        root, model_dir, scene_name, display_table_as_world_space_origin=True
-    ):
-        scene_info = DataLoadUtil.load_scene_info(root, scene_name)
-        target_name = scene_info["target_name"]
-        transformation = scene_info[target_name]
-        if display_table_as_world_space_origin:
-            location = transformation["location"] - DataLoadUtil.get_display_table_top(
-                root, scene_name
-            )
-        else:
-            location = transformation["location"]
-        rotation_euler = transformation["rotation_euler"]
-        pose_mat = trimesh.transformations.euler_matrix(*rotation_euler)
-        pose_mat[:3, 3] = location
-
-        mesh = DataLoadUtil.load_mesh_at(model_dir, target_name, pose_mat)
-        mesh_dir = os.path.join(root, scene_name, "mesh")
-        if not os.path.exists(mesh_dir):
-            os.makedirs(mesh_dir)
-        model_path = os.path.join(mesh_dir, "world_target_mesh.obj")
-        mesh.export(model_path)
-
     @staticmethod
     def load_scene_info(root, scene_name):
         scene_info_path = os.path.join(root, scene_name, "scene_info.json")
@@ -113,17 +97,6 @@ class DataLoadUtil:
             target_pts_num_dict = json.load(f)
         return target_pts_num_dict
 
-    @staticmethod
-    def load_target_object_pose(root, scene_name):
-        scene_info = DataLoadUtil.load_scene_info(root, scene_name)
-        target_name = scene_info["target_name"]
-        transformation = scene_info[target_name]
-        location = transformation["location"]
-        rotation_euler = transformation["rotation_euler"]
-        pose_mat = trimesh.transformations.euler_matrix(*rotation_euler)
-        pose_mat[:3, 3] = location
-        return pose_mat
-
     @staticmethod
     def load_depth(path, min_depth=0.01, max_depth=5.0, binocular=False):
 
@@ -161,8 +134,8 @@ class DataLoadUtil:
         if binocular and not left_only:
 
             def clean_mask(mask_image):
-                green = [0, 255, 0, 255]
-                red = [255, 0, 0, 255]
+                green = [0, 255, 0]
+                red = [255, 0, 0]
                 threshold = 2
                 mask_image = np.where(
                     np.abs(mask_image - green) <= threshold, green, mask_image
@@ -194,30 +167,31 @@ class DataLoadUtil:
             return mask_image
         
     @staticmethod
-    def load_normal(path, binocular=False, left_only=False):
+    def load_normal(path, binocular=False, left_only=False, file_type="exr"):
         if binocular and not left_only:
             normal_path_L = os.path.join(
-                os.path.dirname(path), "normal", os.path.basename(path) + "_L.png"
+                os.path.dirname(path), "normal", os.path.basename(path) + f"_L.{file_type}"
             )
-            normal_image_L = cv2.imread(normal_path_L, cv2.IMREAD_COLOR)
+            normal_image_L = DataLoadUtil.load_exr_image(normal_path_L)
+            
             normal_path_R = os.path.join(
-                os.path.dirname(path), "normal", os.path.basename(path) + "_R.png"
+                os.path.dirname(path), "normal", os.path.basename(path) + f"_R.{file_type}"
             )
-            normal_image_R = cv2.imread(normal_path_R, cv2.IMREAD_COLOR)
-            normalized_normal_image_L = normal_image_L / 255.0 * 2.0 - 1.0
-            normalized_normal_image_R = normal_image_R / 255.0 * 2.0 - 1.0
+            normal_image_R = DataLoadUtil.load_exr_image(normal_path_R)
+            normalized_normal_image_L = normal_image_L * 2.0 - 1.0
+            normalized_normal_image_R = normal_image_R * 2.0 - 1.0
             return normalized_normal_image_L, normalized_normal_image_R
         else:
             if binocular and left_only:
                 normal_path = os.path.join(
-                    os.path.dirname(path), "normal", os.path.basename(path) + "_L.png"
+                    os.path.dirname(path), "normal", os.path.basename(path) + f"_L.{file_type}"
                 )
             else:
                 normal_path = os.path.join(
-                    os.path.dirname(path), "normal", os.path.basename(path) + ".png"
+                    os.path.dirname(path), "normal", os.path.basename(path) + f".{file_type}"
                 )
-            normal_image = cv2.imread(normal_path, cv2.IMREAD_COLOR)
-            normalized_normal_image = normal_image / 255.0 * 2.0 - 1.0
+            normal_image = DataLoadUtil.load_exr_image(normal_path)
+            normalized_normal_image = normal_image * 2.0 - 1.0
             return normalized_normal_image
 
     @staticmethod
@@ -227,20 +201,26 @@ class DataLoadUtil:
         return label_data
 
     @staticmethod
-    def load_rgb(path):
-        rgb_path = os.path.join(
-            os.path.dirname(path), "rgb", os.path.basename(path) + ".png"
-        )
-        rgb_image = cv2.imread(rgb_path, cv2.IMREAD_COLOR)
-        return rgb_image
-
-    @staticmethod
-    def load_from_preprocessed_pts(path):
+    def load_from_preprocessed_pts(path, file_type="npy"):
         npy_path = os.path.join(
-            os.path.dirname(path), "points", os.path.basename(path) + ".npy"
+            os.path.dirname(path), "pts", os.path.basename(path) + "." + file_type
         )
-        pts = np.load(npy_path)
+        if file_type == "txt":
+            pts = np.loadtxt(npy_path)
+        else:
+            pts = np.load(npy_path)
         return pts
+    
+    @staticmethod
+    def load_from_preprocessed_nrm(path, file_type="npy"):
+        npy_path = os.path.join(
+            os.path.dirname(path), "nrm", os.path.basename(path) + "." + file_type
+        )
+        if file_type == "txt":
+            nrm = np.loadtxt(npy_path)
+        else:
+            nrm = np.load(npy_path)
+        return nrm
 
     @staticmethod
     def cam_pose_transformation(cam_pose_before):
@@ -260,11 +240,12 @@ class DataLoadUtil:
             label_data = json.load(f)
         cam_to_world = np.asarray(label_data["extrinsic"])
         cam_to_world = DataLoadUtil.cam_pose_transformation(cam_to_world)
-        world_to_display_table = np.eye(4)
-        world_to_display_table[:3, 3] = -DataLoadUtil.get_display_table_top(
-            root_dir, scene_name
-        )
+        
         if display_table_as_world_space_origin:
+            world_to_display_table = np.eye(4)
+            world_to_display_table[:3, 3] = -DataLoadUtil.get_display_table_top(
+                root_dir, scene_name
+            )
             cam_to_world = np.dot(world_to_display_table, cam_to_world)
         cam_intrinsic = np.asarray(label_data["intrinsic"])
         cam_info = {

utils/pts.py

@@ -1,22 +1,32 @@
 import numpy as np
 import open3d as o3d
 import torch
-from scipy.spatial import cKDTree
 
 class PtsUtil:
 
     @staticmethod
-    def voxel_downsample_point_cloud(point_cloud, voxel_size=0.005):
-        o3d_pc = o3d.geometry.PointCloud()
-        o3d_pc.points = o3d.utility.Vector3dVector(point_cloud)
-        downsampled_pc = o3d_pc.voxel_down_sample(voxel_size)
-        return np.asarray(downsampled_pc.points)
-    
+    def voxel_downsample_point_cloud(point_cloud, voxel_size=0.005, require_idx=False):
+        voxel_indices = np.floor(point_cloud / voxel_size).astype(np.int32)
+        if require_idx:
+            _, inverse, counts = np.unique(voxel_indices, axis=0, return_inverse=True, return_counts=True)
+            idx_sort = np.argsort(inverse)
+            idx_unique = idx_sort[np.cumsum(counts)-counts]
+            downsampled_points = point_cloud[idx_unique]
+            return downsampled_points, idx_unique
+        else:
+            unique_voxels = np.unique(voxel_indices, axis=0, return_inverse=True)
+            return unique_voxels[0]*voxel_size
+        
     @staticmethod
-    def transform_point_cloud(points, pose_mat):
-        points_h = np.concatenate([points, np.ones((points.shape[0], 1))], axis=1)
-        points_h = np.dot(pose_mat, points_h.T).T
-        return points_h[:, :3]
+    def voxel_downsample_point_cloud_random(point_cloud, voxel_size=0.005, require_idx=False):
+        voxel_indices = np.floor(point_cloud / voxel_size).astype(np.int32)
+        unique_voxels, inverse, counts = np.unique(voxel_indices, axis=0, return_inverse=True, return_counts=True)
+        idx_sort = np.argsort(inverse)
+        idx_unique = idx_sort[np.cumsum(counts)-counts]
+        downsampled_points = point_cloud[idx_unique]
+        if require_idx:
+            return downsampled_points, inverse
+        return downsampled_points
     
     @staticmethod
     def random_downsample_point_cloud(point_cloud, num_points, require_idx=False):
@@ -29,6 +39,27 @@ class PtsUtil:
             return point_cloud[idx], idx
         return point_cloud[idx]
     
+    @staticmethod
+    def fps_downsample_point_cloud(point_cloud, num_points, require_idx=False):
+        N = point_cloud.shape[0]
+        mask = np.zeros(N, dtype=bool)
+
+        sampled_indices = np.zeros(num_points, dtype=int)
+        sampled_indices[0] = np.random.randint(0, N)
+        distances = np.linalg.norm(point_cloud - point_cloud[sampled_indices[0]], axis=1)
+        for i in range(1, num_points):
+            farthest_index = np.argmax(distances)
+            sampled_indices[i] = farthest_index
+            mask[farthest_index] = True
+
+            new_distances = np.linalg.norm(point_cloud - point_cloud[farthest_index], axis=1)
+            distances = np.minimum(distances, new_distances)
+        
+        sampled_points = point_cloud[sampled_indices]
+        if require_idx:
+            return sampled_points, sampled_indices
+        return sampled_points
+        
     @staticmethod
     def random_downsample_point_cloud_tensor(point_cloud, num_points):
         idx = torch.randint(0, len(point_cloud), (num_points,))
@@ -40,6 +71,12 @@ class PtsUtil:
         unique_voxels = np.unique(voxel_indices, axis=0, return_inverse=True)
         return unique_voxels
 
+    @staticmethod
+    def transform_point_cloud(points, pose_mat):
+        points_h = np.concatenate([points, np.ones((points.shape[0], 1))], axis=1)
+        points_h = np.dot(pose_mat, points_h.T).T
+        return points_h[:, :3]
+    
     @staticmethod
     def get_overlapping_points(point_cloud_L, point_cloud_R, voxel_size=0.005, require_idx=False):
         voxels_L, indices_L = PtsUtil.voxelize_points(point_cloud_L, voxel_size)
@@ -57,36 +94,24 @@ class PtsUtil:
         return overlapping_points
     
     @staticmethod
-    def new_filter_points(points, normals, cam_pose, theta=75, require_idx=False):
-        camera_axis = -cam_pose[:3, 2] 
+    def filter_points(points, normals, cam_pose, theta_limit=45, z_range=(0.2, 0.45)):
+        
+        """ filter with normal """ 
         normals_normalized = normals / np.linalg.norm(normals, axis=1, keepdims=True)
-        cos_theta = np.dot(normals_normalized, camera_axis)
-        theta_rad = np.deg2rad(theta)
-        idx = cos_theta > np.cos(theta_rad)
-        filtered_points= points[idx]
-        if require_idx:
-            return filtered_points, idx
-        return filtered_points
-    
-    @staticmethod
-    def filter_points(points, points_normals, cam_pose,  voxel_size=0.002, theta=45, z_range=(0.2, 0.45)):
+        cos_theta = np.dot(normals_normalized, np.array([0, 0, 1]))
+        theta = np.arccos(cos_theta) * 180 / np.pi
+        idx = theta < theta_limit
+        filtered_sampled_points = points[idx]
+        filtered_normals = normals[idx]
         
         """ filter with z range """
-        points_cam = PtsUtil.transform_point_cloud(points, np.linalg.inv(cam_pose))
+        points_cam = PtsUtil.transform_point_cloud(filtered_sampled_points, np.linalg.inv(cam_pose))
         idx = (points_cam[:, 2] > z_range[0]) & (points_cam[:, 2] < z_range[1])
-        z_filtered_points = points[idx]
-        
-        """ filter with normal """
-        sampled_points = PtsUtil.voxel_downsample_point_cloud(z_filtered_points, voxel_size)
-        kdtree = cKDTree(points_normals[:,:3])
-        _, indices = kdtree.query(sampled_points)
-        nearest_points = points_normals[indices]
-
-        normals = nearest_points[:, 3:]
-        camera_axis = -cam_pose[:3, 2] 
-        normals_normalized = normals / np.linalg.norm(normals, axis=1, keepdims=True)
-        cos_theta = np.dot(normals_normalized, camera_axis)
-        theta_rad = np.deg2rad(theta)
-        idx = cos_theta > np.cos(theta_rad)
-        filtered_sampled_points= sampled_points[idx] 
-        return filtered_sampled_points[:, :3]
+        z_filtered_points = filtered_sampled_points[idx]
+        z_filtered_normals = filtered_normals[idx]
+        return z_filtered_points[:, :3], z_filtered_normals
+    
+    @staticmethod
+    def point_to_hash(point, voxel_size):
+        return tuple(np.floor(point / voxel_size).astype(int))
+    

utils/reconstruction.py

@@ -3,48 +3,46 @@ from scipy.spatial import cKDTree
 from utils.pts import PtsUtil
 
 class ReconstructionUtil:
-    
+
     @staticmethod
     def compute_coverage_rate(target_point_cloud, combined_point_cloud, threshold=0.01):
         kdtree = cKDTree(combined_point_cloud)
         distances, _ = kdtree.query(target_point_cloud)
-        covered_points = np.sum(distances < threshold*2)
-        coverage_rate = covered_points / target_point_cloud.shape[0]
-        return coverage_rate
+        covered_points_num = np.sum(distances < threshold*2)
+        coverage_rate = covered_points_num / target_point_cloud.shape[0]
+        return coverage_rate, covered_points_num
     
     @staticmethod
-    def compute_overlap_rate(new_point_cloud, combined_point_cloud, threshold=0.01):
+    def compute_coverage_rate_with_normal(target_point_cloud, combined_point_cloud, target_normal, combined_normal, threshold=0.01, normal_threshold=0.1):
+        kdtree = cKDTree(combined_point_cloud)
+        distances, indices = kdtree.query(target_point_cloud)
+        is_covered_by_distance = distances < threshold*2
+        normal_dots = np.einsum('ij,ij->i', target_normal, combined_normal[indices])
+        is_covered_by_normal = normal_dots > normal_threshold
+
+        pts_nrm_target = np.hstack([target_point_cloud, target_normal])
+        np.savetxt("pts_nrm_target.txt",  pts_nrm_target)
+        pts_nrm_combined = np.hstack([combined_point_cloud, combined_normal])
+        np.savetxt("pts_nrm_combined.txt", pts_nrm_combined)
+        import ipdb; ipdb.set_trace()
+        covered_points_num = np.sum(is_covered_by_distance & is_covered_by_normal)
+        coverage_rate = covered_points_num / target_point_cloud.shape[0]
+
+        return coverage_rate, covered_points_num
+        
+    
+    @staticmethod
+    def check_overlap(new_point_cloud, combined_point_cloud, overlap_area_threshold=25, voxel_size=0.01, require_new_added_pts_num=False):
         kdtree = cKDTree(combined_point_cloud)
         distances, _ = kdtree.query(new_point_cloud)
-        overlapping_points = np.sum(distances < threshold)
-        if new_point_cloud.shape[0] == 0:
-            overlap_rate = 0
-        else:
-            overlap_rate = overlapping_points / new_point_cloud.shape[0]
-        return overlap_rate
-    
-    @staticmethod
-    def combine_point_with_view_sequence(point_list, view_sequence):
-        selected_views = []
-        for view_index, _ in view_sequence:
-            selected_views.append(point_list[view_index])
-        return np.vstack(selected_views)
-    
-    @staticmethod 
-    def compute_next_view_coverage_list(views, combined_point_cloud, target_point_cloud, threshold=0.01):
-        best_view = None
-        best_coverage_increase = -1
-        current_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, combined_point_cloud, threshold)
-        
-        for view_index, view in enumerate(views):
-            candidate_views = combined_point_cloud + [view]
-            down_sampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(candidate_views, threshold)
-            new_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, down_sampled_combined_point_cloud, threshold)
-            coverage_increase = new_coverage - current_coverage
-            if coverage_increase > best_coverage_increase:
-                best_coverage_increase = coverage_increase
-                best_view = view_index
-        return best_view, best_coverage_increase
+        overlapping_points_num = np.sum(distances < voxel_size*2)
+        cm = 0.01
+        voxel_size_cm = voxel_size / cm
+        overlap_area = overlapping_points_num * voxel_size_cm * voxel_size_cm
+        if require_new_added_pts_num:
+            return overlap_area > overlap_area_threshold, len(new_point_cloud)-np.sum(distances < voxel_size*1.2)
+        return overlap_area > overlap_area_threshold
+
 
     @staticmethod
     def get_new_added_points(old_combined_pts, new_pts, threshold=0.005):
@@ -59,55 +57,72 @@ class ReconstructionUtil:
         return new_added_points
     
     @staticmethod
-    def compute_next_best_view_sequence_with_overlap(target_point_cloud, point_cloud_list, scan_points_indices_list, threshold=0.01, soft_overlap_threshold=0.5, hard_overlap_threshold=0.7, init_view = 0, scan_points_threshold=5, status_info=None):
-        selected_views = [point_cloud_list[init_view]]
-        combined_point_cloud = np.vstack(selected_views)
+    def compute_next_best_view_sequence(target_point_cloud, point_cloud_list, scan_points_indices_list, threshold=0.01, overlap_area_threshold=25, init_view = 0, scan_points_threshold=5, status_info=None):
+        selected_views = [init_view]
+        combined_point_cloud = point_cloud_list[init_view]
         history_indices = [scan_points_indices_list[init_view]]
-        down_sampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_point_cloud,threshold)
-        new_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, down_sampled_combined_point_cloud, threshold)
+        
+        max_rec_pts = np.vstack(point_cloud_list)
+        downsampled_max_rec_pts = PtsUtil.voxel_downsample_point_cloud(max_rec_pts, threshold)
+        combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_point_cloud, threshold)
+        max_rec_pts_num = downsampled_max_rec_pts.shape[0]
+        max_real_rec_pts_coverage, _ = ReconstructionUtil.compute_coverage_rate(target_point_cloud, downsampled_max_rec_pts, threshold)
+        
+        new_coverage, new_covered_num = ReconstructionUtil.compute_coverage_rate(downsampled_max_rec_pts, combined_point_cloud, threshold)
         current_coverage = new_coverage
+        current_covered_num = new_covered_num
+        
         remaining_views = list(range(len(point_cloud_list)))
         view_sequence = [(init_view, current_coverage)]
         cnt_processed_view = 0
         remaining_views.remove(init_view)
-
+        curr_rec_pts_num = combined_point_cloud.shape[0]
+        drop_output_ratio = 0.4
+        
+        import time
         while remaining_views:
             best_view = None
             best_coverage_increase = -1
+            best_combined_point_cloud = None
+            best_covered_num = 0
         
             for view_index in remaining_views:
+                if np.random.rand() < drop_output_ratio:
+                    continue
                 if point_cloud_list[view_index].shape[0] == 0:
                     continue
-
                 if selected_views:
                     new_scan_points_indices = scan_points_indices_list[view_index]
-                    
                     if not ReconstructionUtil.check_scan_points_overlap(history_indices, new_scan_points_indices, scan_points_threshold):
-                        overlap_threshold = hard_overlap_threshold
+                        curr_overlap_area_threshold = overlap_area_threshold
                     else:
-                        overlap_threshold = soft_overlap_threshold
+                        curr_overlap_area_threshold = overlap_area_threshold * 0.5
                         
-                    combined_old_point_cloud = np.vstack(selected_views)
-                    down_sampled_old_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_old_point_cloud,threshold)
-                    down_sampled_new_view_point_cloud = PtsUtil.voxel_downsample_point_cloud(point_cloud_list[view_index],threshold)
-                    overlap_rate = ReconstructionUtil.compute_overlap_rate(down_sampled_new_view_point_cloud,down_sampled_old_point_cloud, threshold)
-                    if overlap_rate < overlap_threshold:
+                    if not ReconstructionUtil.check_overlap(point_cloud_list[view_index], combined_point_cloud, overlap_area_threshold = curr_overlap_area_threshold, voxel_size=threshold):
                         continue
-                    
-                candidate_views = selected_views + [point_cloud_list[view_index]]
-                combined_point_cloud = np.vstack(candidate_views)
-                down_sampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(combined_point_cloud,threshold)
-                new_coverage = ReconstructionUtil.compute_coverage_rate(target_point_cloud, down_sampled_combined_point_cloud, threshold)
+                
+                new_combined_point_cloud = np.vstack([combined_point_cloud, point_cloud_list[view_index]])
+                new_downsampled_combined_point_cloud = PtsUtil.voxel_downsample_point_cloud(new_combined_point_cloud,threshold)
+                new_coverage, new_covered_num = ReconstructionUtil.compute_coverage_rate(downsampled_max_rec_pts, new_downsampled_combined_point_cloud, threshold)
                 coverage_increase = new_coverage - current_coverage
                 if coverage_increase > best_coverage_increase:
                     best_coverage_increase = coverage_increase
                     best_view = view_index
+                    best_covered_num = new_covered_num
+                    best_combined_point_cloud = new_downsampled_combined_point_cloud
                     
             
             if best_view is not None:
-                if best_coverage_increase <=3e-3:
+                if best_coverage_increase <=1e-3 or best_covered_num - current_covered_num <= 5:
                     break
-                selected_views.append(point_cloud_list[best_view])
+                
+                selected_views.append(best_view)
+                best_rec_pts_num = best_combined_point_cloud.shape[0]
+                print(f"Current rec pts num: {curr_rec_pts_num}, Best rec pts num: {best_rec_pts_num}, Best cover pts: {best_covered_num}, Max rec pts num: {max_rec_pts_num}")
+                print(f"Current coverage: {current_coverage+best_coverage_increase}, Best coverage increase: {best_coverage_increase}, Max Real coverage: {max_real_rec_pts_coverage}")
+                current_covered_num = best_covered_num
+                curr_rec_pts_num = best_rec_pts_num
+                combined_point_cloud = best_combined_point_cloud
                 remaining_views.remove(best_view)
                 history_indices.append(scan_points_indices_list[best_view])
                 current_coverage += best_coverage_increase
@@ -128,7 +143,102 @@ class ReconstructionUtil:
             app_name = status_info["app_name"]
             runner_name = status_info["runner_name"]
             sm.set_progress(app_name, runner_name, "processed view", len(point_cloud_list), len(point_cloud_list))
-        return view_sequence, remaining_views, down_sampled_combined_point_cloud
+        return view_sequence, remaining_views, combined_point_cloud
+    
+    @staticmethod
+    def compute_next_best_view_sequence_with_normal(target_point_cloud, target_normal, point_cloud_list, normal_list, scan_points_indices_list, threshold=0.01, overlap_area_threshold=25, init_view = 0, scan_points_threshold=5, status_info=None):
+        selected_views = [init_view]
+        combined_point_cloud = point_cloud_list[init_view]
+        combined_normal = normal_list[init_view]
+        history_indices = [scan_points_indices_list[init_view]]
+        
+        max_rec_pts = np.vstack(point_cloud_list)
+        max_rec_nrm = np.vstack(normal_list)
+        downsampled_max_rec_pts, idx = PtsUtil.voxel_downsample_point_cloud(max_rec_pts, threshold, require_idx=True)
+        downsampled_max_rec_nrm = max_rec_nrm[idx]  
+        max_rec_pts_num = downsampled_max_rec_pts.shape[0]
+        try:
+            max_real_rec_pts_coverage, _ = ReconstructionUtil.compute_coverage_rate_with_normal(target_point_cloud, downsampled_max_rec_pts, target_normal, downsampled_max_rec_nrm, threshold)
+        except:
+            import ipdb; ipdb.set_trace()
+        
+        new_coverage, new_covered_num = ReconstructionUtil.compute_coverage_rate_with_normal(downsampled_max_rec_pts, combined_point_cloud, downsampled_max_rec_nrm, combined_normal, threshold)
+        current_coverage = new_coverage
+        current_covered_num = new_covered_num
+        
+        remaining_views = list(range(len(point_cloud_list)))
+        view_sequence = [(init_view, current_coverage)]
+        cnt_processed_view = 0
+        remaining_views.remove(init_view)
+        curr_rec_pts_num = combined_point_cloud.shape[0]
+        
+        while remaining_views:
+            best_view = None
+            best_coverage_increase = -1
+            best_combined_point_cloud = None
+            best_combined_normal = None
+            best_covered_num = 0
+        
+            for view_index in remaining_views:
+                if point_cloud_list[view_index].shape[0] == 0:
+                    continue
+                if selected_views:
+                    new_scan_points_indices = scan_points_indices_list[view_index]
+                    if not ReconstructionUtil.check_scan_points_overlap(history_indices, new_scan_points_indices, scan_points_threshold):
+                        curr_overlap_area_threshold = overlap_area_threshold
+                    else:
+                        curr_overlap_area_threshold = overlap_area_threshold * 0.5
+                        
+                    if not ReconstructionUtil.check_overlap(point_cloud_list[view_index], combined_point_cloud, overlap_area_threshold = curr_overlap_area_threshold, voxel_size=threshold):
+                        continue
+                
+                new_combined_point_cloud = np.vstack([combined_point_cloud, point_cloud_list[view_index]])
+                new_combined_normal = np.vstack([combined_normal, normal_list[view_index]])
+                new_downsampled_combined_point_cloud, idx = PtsUtil.voxel_downsample_point_cloud(new_combined_point_cloud,threshold, require_idx=True)
+                new_downsampled_combined_normal = new_combined_normal[idx]
+                new_coverage, new_covered_num = ReconstructionUtil.compute_coverage_rate_with_normal(downsampled_max_rec_pts, new_downsampled_combined_point_cloud, downsampled_max_rec_nrm, new_downsampled_combined_normal, threshold)
+                coverage_increase = new_coverage - current_coverage
+                if coverage_increase > best_coverage_increase:
+                    best_coverage_increase = coverage_increase
+                    best_view = view_index
+                    best_covered_num = new_covered_num
+                    best_combined_point_cloud = new_downsampled_combined_point_cloud
+                    best_combined_normal = new_downsampled_combined_normal
+                    
+            
+            if best_view is not None:
+                if best_coverage_increase <=1e-3 or best_covered_num - current_covered_num <= 5:
+                    break
+                
+                selected_views.append(best_view)
+                best_rec_pts_num = best_combined_point_cloud.shape[0]
+                print(f"Current rec pts num: {curr_rec_pts_num}, Best rec pts num: {best_rec_pts_num}, Best cover pts: {best_covered_num}, Max rec pts num: {max_rec_pts_num}")
+                print(f"Current coverage: {current_coverage}, Best coverage increase: {best_coverage_increase}, Max Real coverage: {max_real_rec_pts_coverage}")
+                current_covered_num = best_covered_num
+                curr_rec_pts_num = best_rec_pts_num
+                combined_point_cloud = best_combined_point_cloud
+                combined_normal = best_combined_normal
+                remaining_views.remove(best_view)
+                history_indices.append(scan_points_indices_list[best_view])
+                current_coverage += best_coverage_increase
+                cnt_processed_view += 1
+                if status_info is not None:
+                    sm = status_info["status_manager"]
+                    app_name = status_info["app_name"]
+                    runner_name = status_info["runner_name"]
+                    sm.set_status(app_name, runner_name, "current coverage", current_coverage)
+                    sm.set_progress(app_name, runner_name, "processed view", cnt_processed_view, len(point_cloud_list))
+            
+                view_sequence.append((best_view, current_coverage))
+                
+            else:
+                break
+        if status_info is not None:
+            sm = status_info["status_manager"]
+            app_name = status_info["app_name"]
+            runner_name = status_info["runner_name"]
+            sm.set_progress(app_name, runner_name, "processed view", len(point_cloud_list), len(point_cloud_list))
+        return view_sequence, remaining_views, combined_point_cloud
 
     
     @staticmethod
@@ -147,18 +257,6 @@ class ReconstructionUtil:
             attempts += 1
         return points
     
-    @staticmethod
-    def compute_covered_scan_points(scan_points, point_cloud, threshold=0.01):
-        
-        tree = cKDTree(point_cloud)
-        covered_points = []
-        indices = []
-        for i, scan_point in enumerate(scan_points):
-            if tree.query_ball_point(scan_point, threshold):
-                covered_points.append(scan_point)
-                indices.append(i)
-        return covered_points, indices
-    
     @staticmethod
     def check_scan_points_overlap(history_indices, indices2, threshold=5):
         for indices1 in history_indices:

utils/render.py

@@ -1,16 +1,75 @@
 
 import os
 import json
+import time
 import subprocess
 import tempfile
 import shutil
+import numpy as np
 from utils.data_load import DataLoadUtil
 from utils.reconstruction import ReconstructionUtil
 from utils.pts import PtsUtil
 class RenderUtil:
+    target_mask_label = (0, 255, 0)
+    display_table_mask_label = (0, 0, 255)
+    random_downsample_N = 32768
+    min_z = 0.2
+    max_z = 0.5
 
     @staticmethod
-    def render_pts(cam_pose, scene_path, script_path, model_points_normals, voxel_threshold=0.005, filter_degree=75, nO_to_nL_pose=None, require_full_scene=False):
+    def get_world_points_and_normal(depth, mask, normal, cam_intrinsic, cam_extrinsic, random_downsample_N):
+        z = depth[mask]
+        i, j = np.nonzero(mask)
+        x = (j - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
+        y = (i - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
+
+        points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
+        normal_camera = normal[mask].reshape(-1, 3)
+        sampled_target_points, idx = PtsUtil.random_downsample_point_cloud(
+                    points_camera, random_downsample_N, require_idx=True
+                )
+        if len(sampled_target_points) == 0:
+            return np.zeros((0, 3)), np.zeros((0, 3))
+        sampled_normal_camera  = normal_camera[idx]
+
+        points_camera_aug = np.concatenate((sampled_target_points, np.ones((sampled_target_points.shape[0], 1))), axis=-1)
+        points_camera_world = np.dot(cam_extrinsic, points_camera_aug.T).T[:, :3]
+        
+        return points_camera_world, sampled_normal_camera
+    
+    @staticmethod
+    def get_world_points(depth, mask, cam_intrinsic, cam_extrinsic, random_downsample_N):
+        z = depth[mask]
+        i, j = np.nonzero(mask)
+        x = (j - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
+        y = (i - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
+
+        points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
+        sampled_target_points = PtsUtil.random_downsample_point_cloud(
+                    points_camera, random_downsample_N
+                )
+        points_camera_aug = np.concatenate((sampled_target_points, np.ones((sampled_target_points.shape[0], 1))), axis=-1)
+        points_camera_world = np.dot(cam_extrinsic, points_camera_aug.T).T[:, :3]
+        
+        return points_camera_world
+    
+    @staticmethod
+    def get_scan_points_indices(scan_points, mask, display_table_mask_label, cam_intrinsic, cam_extrinsic):
+        scan_points_homogeneous = np.hstack((scan_points, np.ones((scan_points.shape[0], 1))))
+        points_camera = np.dot(np.linalg.inv(cam_extrinsic), scan_points_homogeneous.T).T[:, :3]
+        points_image_homogeneous = np.dot(cam_intrinsic, points_camera.T).T
+        points_image_homogeneous /= points_image_homogeneous[:, 2:]
+        pixel_x = points_image_homogeneous[:, 0].astype(int)
+        pixel_y = points_image_homogeneous[:, 1].astype(int)
+        h, w = mask.shape[:2]
+        valid_indices = (pixel_x >= 0) & (pixel_x < w) & (pixel_y >= 0) & (pixel_y < h)
+        mask_colors = mask[pixel_y[valid_indices], pixel_x[valid_indices]]
+        selected_points_indices = np.where((mask_colors == display_table_mask_label).all(axis=-1))[0]
+        selected_points_indices = np.where(valid_indices)[0][selected_points_indices]
+        return selected_points_indices
+
+    @staticmethod
+    def render_pts(cam_pose, scene_path, script_path, scan_points, voxel_threshold=0.005, filter_degree=75, nO_to_nL_pose=None, require_full_scene=False):
         
         nO_to_world_pose = DataLoadUtil.get_real_cam_O_from_cam_L(cam_pose, nO_to_nL_pose, scene_path=scene_path)
 
@@ -26,27 +85,51 @@ class RenderUtil:
             with open(params_data_path, 'w') as f:
                 json.dump(params, f)
             result = subprocess.run([
-                'blender', '-b', '-P', script_path, '--', temp_dir
+                '/home/hofee/blender-4.0.2-linux-x64/blender', '-b', '-P', script_path, '--', temp_dir
             ], capture_output=True, text=True)
-            if result.returncode != 0:
-                print("Blender script failed:")
-                print(result.stderr)
-                return None
+            # print(result)
             path = os.path.join(temp_dir, "tmp")
-            point_cloud = DataLoadUtil.get_target_point_cloud_world_from_path(path, binocular=True)
-            cam_params = DataLoadUtil.load_cam_info(path, binocular=True)
-            
-            ''' TODO: old code: filter_points api is changed, need to update the code '''
-            filtered_point_cloud = PtsUtil.filter_points(point_cloud, model_points_normals, cam_pose=cam_params["cam_to_world"], voxel_size=voxel_threshold, theta=filter_degree)
-            full_scene_point_cloud = None
-            if require_full_scene:
-                depth_L, depth_R = DataLoadUtil.load_depth(path, cam_params['near_plane'], cam_params['far_plane'], binocular=True)
-                point_cloud_L = DataLoadUtil.get_point_cloud(depth_L, cam_params['cam_intrinsic'], cam_params['cam_to_world'])['points_world']
-                point_cloud_R = DataLoadUtil.get_point_cloud(depth_R, cam_params['cam_intrinsic'], cam_params['cam_to_world_R'])['points_world']
-            
-                point_cloud_L = PtsUtil.random_downsample_point_cloud(point_cloud_L, 65536)
-                point_cloud_R = PtsUtil.random_downsample_point_cloud(point_cloud_R, 65536)
-                full_scene_point_cloud = PtsUtil.get_overlapping_points(point_cloud_L, point_cloud_R)
+            cam_info = DataLoadUtil.load_cam_info(path, binocular=True)
+            depth_L, depth_R = DataLoadUtil.load_depth(
+                    path, cam_info["near_plane"], 
+                    cam_info["far_plane"], 
+                    binocular=True
+                )
+            mask_L, mask_R = DataLoadUtil.load_seg(path, binocular=True)
+            normal_L = DataLoadUtil.load_normal(path, binocular=True, left_only=True)
+            ''' target points '''
+            mask_img_L = mask_L
+            mask_img_R = mask_R
+
+            target_mask_img_L = (mask_L == RenderUtil.target_mask_label).all(axis=-1)
+            target_mask_img_R = (mask_R == RenderUtil.target_mask_label).all(axis=-1)
 
             
-            return filtered_point_cloud, full_scene_point_cloud
+            sampled_target_points_L, sampled_target_normal_L = RenderUtil.get_world_points_and_normal(depth_L,target_mask_img_L,normal_L, cam_info["cam_intrinsic"], cam_info["cam_to_world"],  RenderUtil.random_downsample_N)
+            sampled_target_points_R = RenderUtil.get_world_points(depth_R, target_mask_img_R, cam_info["cam_intrinsic"], cam_info["cam_to_world_R"],  RenderUtil.random_downsample_N  )
+
+
+            has_points = sampled_target_points_L.shape[0] > 0 and sampled_target_points_R.shape[0] > 0
+            if has_points:
+                target_points, overlap_idx = PtsUtil.get_overlapping_points(
+                        sampled_target_points_L, sampled_target_points_R, voxel_threshold, require_idx=True
+                    )
+                sampled_target_normal_L = sampled_target_normal_L[overlap_idx]
+
+            if has_points:
+                has_points = target_points.shape[0] > 0
+
+            if has_points:
+                target_points, target_normals = PtsUtil.filter_points(
+                    target_points, sampled_target_normal_L, cam_info["cam_to_world"], theta_limit = filter_degree, z_range=(RenderUtil.min_z, RenderUtil.max_z)
+                    )
+                
+            
+            scan_points_indices_L = RenderUtil.get_scan_points_indices(scan_points, mask_img_L, RenderUtil.display_table_mask_label, cam_info["cam_intrinsic"], cam_info["cam_to_world"]) 
+            scan_points_indices_R = RenderUtil.get_scan_points_indices(scan_points, mask_img_R, RenderUtil.display_table_mask_label, cam_info["cam_intrinsic"], cam_info["cam_to_world_R"])
+            scan_points_indices = np.intersect1d(scan_points_indices_L, scan_points_indices_R)
+            if not has_points:
+                target_points = np.zeros((0, 3))
+                target_normals = np.zeros((0, 3))
+            #import ipdb; ipdb.set_trace()
+            return target_points, target_normals, scan_points_indices

utils/vis.py

@@ -0,0 +1,195 @@
+import numpy as np
+import matplotlib.pyplot as plt
+import sys
+import os
+import trimesh
+
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+from utils.data_load import DataLoadUtil
+from utils.pts import PtsUtil
+
+class visualizeUtil:
+    
+    @staticmethod
+    def save_all_cam_pos_and_cam_axis(root, scene, output_dir):
+        length = DataLoadUtil.get_scene_seq_length(root, scene)
+        all_cam_pos = []
+        all_cam_axis = []
+        for i in range(length):
+            path = DataLoadUtil.get_path(root, scene, i)
+            cam_info = DataLoadUtil.load_cam_info(path, binocular=True)
+            cam_pose = cam_info["cam_to_world"]
+            cam_pos = cam_pose[:3, 3]
+            cam_axis = cam_pose[:3, 2] 
+            
+            num_samples = 10
+            sample_points = [cam_pos + 0.02*t * cam_axis for t in range(num_samples)]
+            sample_points = np.array(sample_points)
+        
+            all_cam_pos.append(cam_pos)
+            all_cam_axis.append(sample_points)
+        
+        all_cam_pos = np.array(all_cam_pos)
+        all_cam_axis = np.array(all_cam_axis).reshape(-1, 3)
+        np.savetxt(os.path.join(output_dir, "all_cam_pos.txt"), all_cam_pos)
+        np.savetxt(os.path.join(output_dir, "all_cam_axis.txt"), all_cam_axis)
+        
+    @staticmethod
+    def save_all_combined_pts(root, scene, output_dir):
+        length = DataLoadUtil.get_scene_seq_length(root, scene)
+        all_combined_pts = []   
+        for i in range(length):
+            path = DataLoadUtil.get_path(root, scene, i)
+            pts = DataLoadUtil.load_from_preprocessed_pts(path,"npy")
+            if pts.shape[0] == 0:
+                continue
+            all_combined_pts.append(pts)
+        all_combined_pts = np.vstack(all_combined_pts)
+        downsampled_all_pts = PtsUtil.voxel_downsample_point_cloud(all_combined_pts, 0.001)
+        np.savetxt(os.path.join(output_dir, "all_combined_pts.txt"), downsampled_all_pts)
+
+    @staticmethod
+    def save_seq_cam_pos_and_cam_axis(root, scene, frame_idx_list, output_dir):
+        all_cam_pos = []
+        all_cam_axis = []
+        for i in frame_idx_list:
+            path = DataLoadUtil.get_path(root, scene, i)
+            cam_info = DataLoadUtil.load_cam_info(path, binocular=True)
+            cam_pose = cam_info["cam_to_world"]
+            cam_pos = cam_pose[:3, 3]
+            cam_axis = cam_pose[:3, 2] 
+            
+            num_samples = 10
+            sample_points = [cam_pos + 0.02*t * cam_axis for t in range(num_samples)]
+            sample_points = np.array(sample_points)
+        
+            all_cam_pos.append(cam_pos)
+            all_cam_axis.append(sample_points)
+        
+        all_cam_pos = np.array(all_cam_pos)
+        all_cam_axis = np.array(all_cam_axis).reshape(-1, 3)
+        np.savetxt(os.path.join(output_dir, "seq_cam_pos.txt"), all_cam_pos)
+        np.savetxt(os.path.join(output_dir, "seq_cam_axis.txt"), all_cam_axis)
+        
+    @staticmethod
+    def save_seq_combined_pts(root, scene, frame_idx_list, output_dir):
+        all_combined_pts = []   
+        for i in frame_idx_list:
+            path = DataLoadUtil.get_path(root, scene, i)
+            pts = DataLoadUtil.load_from_preprocessed_pts(path,"npy")
+            if pts.shape[0] == 0:
+                continue
+            all_combined_pts.append(pts)
+        all_combined_pts = np.vstack(all_combined_pts)
+        downsampled_all_pts = PtsUtil.voxel_downsample_point_cloud(all_combined_pts, 0.001)
+        np.savetxt(os.path.join(output_dir, "seq_combined_pts.txt"), downsampled_all_pts)
+        
+    @staticmethod
+    def save_target_mesh_at_world_space(
+        root, model_dir, scene_name, display_table_as_world_space_origin=True
+    ):
+        scene_info = DataLoadUtil.load_scene_info(root, scene_name)
+        target_name = scene_info["target_name"]
+        transformation = scene_info[target_name]
+        if display_table_as_world_space_origin:
+            location = transformation["location"] - DataLoadUtil.get_display_table_top(
+                root, scene_name
+            )
+        else:
+            location = transformation["location"]
+        rotation_euler = transformation["rotation_euler"]
+        pose_mat = trimesh.transformations.euler_matrix(*rotation_euler)
+        pose_mat[:3, 3] = location
+
+        mesh = DataLoadUtil.load_mesh_at(model_dir, target_name, pose_mat)
+        mesh_dir = os.path.join(root, scene_name, "mesh")
+        if not os.path.exists(mesh_dir):
+            os.makedirs(mesh_dir)
+        model_path = os.path.join(mesh_dir, "world_target_mesh.obj")
+        mesh.export(model_path)
+
+    @staticmethod
+    def save_points_and_normals(root, scene, frame_idx, output_dir, binocular=False):
+        target_mask_label = (0, 255, 0, 255)
+        path = DataLoadUtil.get_path(root, scene, frame_idx)
+        cam_info = DataLoadUtil.load_cam_info(path, binocular=binocular, display_table_as_world_space_origin=False)
+        depth = DataLoadUtil.load_depth(
+                path, cam_info["near_plane"], 
+                cam_info["far_plane"], 
+                binocular=binocular,
+            )
+        if isinstance(depth, tuple):
+            depth = depth[0]
+            
+        mask = DataLoadUtil.load_seg(path, binocular=binocular, left_only=True)
+        normal = DataLoadUtil.load_normal(path, binocular=binocular, left_only=True)
+        ''' target points '''
+        if mask is None:
+            target_mask_img = np.ones_like(depth, dtype=bool)
+        else:
+            target_mask_img = (mask == target_mask_label).all(axis=-1)
+        cam_intrinsic = cam_info["cam_intrinsic"]
+        z = depth[target_mask_img]
+        i, j = np.nonzero(target_mask_img)
+        x = (j - cam_intrinsic[0, 2]) * z / cam_intrinsic[0, 0]
+        y = (i - cam_intrinsic[1, 2]) * z / cam_intrinsic[1, 1]
+        
+        random_downsample_N = 1000
+
+        points_camera = np.stack((x, y, z), axis=-1).reshape(-1, 3)
+        normal_camera = normal[target_mask_img].reshape(-1, 3)
+        sampled_target_points, idx = PtsUtil.random_downsample_point_cloud(
+                    points_camera, random_downsample_N, require_idx=True
+                )
+        if len(sampled_target_points) == 0:
+            print("No target points")
+            
+
+        sampled_normal_camera  = normal_camera[idx]
+        sampled_visualized_normal = []
+        sampled_normal_camera[:, 2] = -sampled_normal_camera[:, 2]
+        sampled_normal_camera[:, 1] = -sampled_normal_camera[:, 1]
+        num_samples = 10
+        for i in range(len(sampled_target_points)):
+            sampled_visualized_normal.append([sampled_target_points[i] + 0.02*t * sampled_normal_camera[i] for t in range(num_samples)])
+            
+        sampled_visualized_normal = np.array(sampled_visualized_normal).reshape(-1, 3)
+        np.savetxt(os.path.join(output_dir, "target_pts.txt"), sampled_target_points)
+        np.savetxt(os.path.join(output_dir, "target_normal.txt"), sampled_visualized_normal)
+
+    @staticmethod
+    def save_pts_nrm(root, scene, frame_idx, output_dir, binocular=False):
+        path = DataLoadUtil.get_path(root, scene, frame_idx)
+        pts_world = DataLoadUtil.load_from_preprocessed_pts(path, "npy")
+        nrm_camera = DataLoadUtil.load_from_preprocessed_nrm(path, "npy")
+        cam_info = DataLoadUtil.load_cam_info(path, binocular=binocular)
+        cam_to_world = cam_info["cam_to_world"]
+        nrm_world = nrm_camera @ cam_to_world[:3, :3].T
+        visualized_nrm = []
+        num_samples = 10
+        for i in range(len(pts_world)):
+            for t in range(num_samples):
+                visualized_nrm.append(pts_world[i] - 0.02 * t * nrm_world[i])
+        
+        visualized_nrm = np.array(visualized_nrm)
+        np.savetxt(os.path.join(output_dir, "nrm.txt"), visualized_nrm)
+        np.savetxt(os.path.join(output_dir, "pts.txt"), pts_world)
+    
+    # @staticmethod
+    # def save_
+        
+# ------ Debug ------
+
+if __name__ == "__main__":
+    root = r"C:\Document\Local Project\nbv_rec\nbv_reconstruction\temp"
+    model_dir = r"H:\\AI\\Datasets\\scaled_object_box_meshes"
+    scene = "box"
+    output_dir = r"C:\Document\Local Project\nbv_rec\nbv_reconstruction\test"
+    
+    #visualizeUtil.save_all_cam_pos_and_cam_axis(root, scene, output_dir)
+    # visualizeUtil.save_all_combined_pts(root, scene, output_dir)
+    # visualizeUtil.save_seq_combined_pts(root, scene, [0, 121, 286, 175, 111,366,45,230,232,225,255,17,199,78,60], output_dir)
+    # visualizeUtil.save_seq_cam_pos_and_cam_axis(root, scene, [0, 121, 286, 175, 111,366,45,230,232,225,255,17,199,78,60], output_dir)
+    # visualizeUtil.save_target_mesh_at_world_space(root, model_dir, scene)
+    #visualizeUtil.save_points_and_normals(root, scene,"10", output_dir, binocular=True)
+    visualizeUtil.save_pts_nrm(root, scene, "116", output_dir, binocular=True)