update strong p++

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()

beans/predict_result.py

@@ -0,0 +1,162 @@
+import numpy as np
+from sklearn.cluster import DBSCAN
+
+class PredictResult:
+    def __init__(self, raw_predict_result, input_pts=None, cluster_params=dict(eps=0.5, min_samples=2)):
+        self.input_pts = input_pts
+        self.cluster_params = cluster_params
+        self.sampled_9d_pose = raw_predict_result
+        self.sampled_matrix_pose = self.get_sampled_matrix_pose()
+        self.distance_matrix = self.calculate_distance_matrix()
+        self.clusters = self.get_cluster_result()
+        self.candidate_matrix_poses = self.get_candidate_poses()
+        self.candidate_9d_poses = [np.concatenate((self.matrix_to_rotation_6d_numpy(matrix[:3,:3]), matrix[:3,3].reshape(-1,)), axis=-1) for matrix in self.candidate_matrix_poses]
+        self.cluster_num = len(self.clusters)
+
+    @staticmethod
+    def rotation_6d_to_matrix_numpy(d6):
+        a1, a2 = d6[:3], d6[3:]
+        b1 = a1 / np.linalg.norm(a1)
+        b2 = a2 - np.dot(b1, a2) * b1
+        b2 = b2 / np.linalg.norm(b2)
+        b3 = np.cross(b1, b2)
+        return np.stack((b1, b2, b3), axis=-2)
+    
+    @staticmethod
+    def matrix_to_rotation_6d_numpy(matrix):
+        return np.copy(matrix[:2, :]).reshape((6,))
+    
+    def __str__(self):
+        info = "Predict Result:\n"
+        info += f"  Predicted pose number: {len(self.sampled_9d_pose)}\n"
+        info += f"  Cluster number: {self.cluster_num}\n"
+        for i, cluster in enumerate(self.clusters):
+            info += f"   - Cluster {i} size: {len(cluster)}\n"
+        max_distance = np.max(self.distance_matrix[self.distance_matrix != 0])
+        min_distance = np.min(self.distance_matrix[self.distance_matrix != 0])
+        info += f"  Max distance: {max_distance}\n"
+        info += f"  Min distance: {min_distance}\n"
+        return info
+    
+    def get_sampled_matrix_pose(self):
+        sampled_matrix_pose = []
+        for pose in self.sampled_9d_pose:
+            rotation = pose[:6]
+            translation = pose[6:]
+            pose = self.rotation_6d_to_matrix_numpy(rotation)
+            pose = np.concatenate((pose, translation.reshape(-1, 1)), axis=-1)
+            pose = np.concatenate((pose, np.array([[0, 0, 0, 1]])), axis=-2)
+            sampled_matrix_pose.append(pose)
+        return np.array(sampled_matrix_pose)
+
+    def rotation_distance(self, R1, R2):
+        R = np.dot(R1.T, R2)
+        trace = np.trace(R)
+        angle = np.arccos(np.clip((trace - 1) / 2, -1, 1))
+        return angle
+
+    def calculate_distance_matrix(self):
+        n = len(self.sampled_matrix_pose)
+        dist_matrix = np.zeros((n, n))
+        for i in range(n):
+            for j in range(n):
+                dist_matrix[i, j] = self.rotation_distance(self.sampled_matrix_pose[i][:3, :3], self.sampled_matrix_pose[j][:3, :3])
+        return dist_matrix
+
+    def cluster_rotations(self):
+        clustering = DBSCAN(eps=self.cluster_params['eps'], min_samples=self.cluster_params['min_samples'], metric='precomputed')
+        labels = clustering.fit_predict(self.distance_matrix)
+        return labels
+
+    def get_cluster_result(self):
+        labels = self.cluster_rotations()
+        cluster_num = len(set(labels)) - (1 if -1 in labels else 0)
+        clusters = []
+        for _ in range(cluster_num):
+            clusters.append([])
+        for matrix_pose, label in zip(self.sampled_matrix_pose, labels):
+            if label != -1:
+                clusters[label].append(matrix_pose)
+        clusters.sort(key=len, reverse=True)
+        return clusters
+
+    def get_center_matrix_pose_from_cluster(self, cluster):
+        min_total_distance = float('inf')
+        center_matrix_pose = None
+        
+        for matrix_pose in cluster:
+            total_distance = 0
+            for other_matrix_pose in cluster:
+                rot_distance = self.rotation_distance(matrix_pose[:3, :3], other_matrix_pose[:3, :3])
+                total_distance += rot_distance
+            
+            if total_distance < min_total_distance:
+                min_total_distance = total_distance
+                center_matrix_pose = matrix_pose
+        
+        return center_matrix_pose
+    
+    def get_candidate_poses(self):
+        candidate_poses = []
+        for cluster in self.clusters:
+            candidate_poses.append(self.get_center_matrix_pose_from_cluster(cluster))
+        return candidate_poses
+    
+    def visualize(self):
+        import plotly.graph_objects as go
+        fig = go.Figure()
+        if self.input_pts is not None:
+            fig.add_trace(go.Scatter3d(
+                x=self.input_pts[:, 0], y=self.input_pts[:, 1], z=self.input_pts[:, 2],
+                mode='markers', marker=dict(size=1, color='gray', opacity=0.5), name='Input Points'
+            ))
+        colors = ['aggrnyl', 'agsunset', 'algae', 'amp', 'armyrose', 'balance',
+             'blackbody', 'bluered', 'blues', 'blugrn', 'bluyl', 'brbg']
+        for i, cluster in enumerate(self.clusters):
+            color = colors[i]
+            candidate_pose = self.candidate_matrix_poses[i]
+            origin_candidate = candidate_pose[:3, 3]
+            z_axis_candidate = candidate_pose[:3, 2]
+            for pose in cluster:
+                origin = pose[:3, 3]
+                z_axis = pose[:3, 2]
+                fig.add_trace(go.Cone(
+                    x=[origin[0]], y=[origin[1]], z=[origin[2]],
+                    u=[z_axis[0]], v=[z_axis[1]], w=[z_axis[2]],
+                    colorscale=color,
+                    sizemode="absolute", sizeref=0.05, anchor="tail", showscale=False
+                ))
+            fig.add_trace(go.Cone(
+                x=[origin_candidate[0]], y=[origin_candidate[1]], z=[origin_candidate[2]],
+                u=[z_axis_candidate[0]], v=[z_axis_candidate[1]], w=[z_axis_candidate[2]],
+                colorscale=color,
+                sizemode="absolute", sizeref=0.1, anchor="tail", showscale=False
+            ))
+        
+        fig.update_layout(
+            title="Clustered Poses and Input Points",
+            scene=dict(
+                xaxis_title='X',
+                yaxis_title='Y',
+                zaxis_title='Z'
+            ),
+            margin=dict(l=0, r=0, b=0, t=40),
+            scene_camera=dict(eye=dict(x=1.25, y=1.25, z=1.25))
+        )
+        
+        fig.show()
+
+
+
+if __name__ == "__main__":
+    step = 0
+    raw_predict_result = np.load(f"inference_result_pack/inference_result_pack/{step}/all_pred_pose_9d.npy")
+    input_pts = np.loadtxt(f"inference_result_pack/inference_result_pack/{step}/input_pts.txt")
+    print(raw_predict_result.shape)
+    predict_result = PredictResult(raw_predict_result, input_pts, cluster_params=dict(eps=0.25, min_samples=3))
+    print(predict_result)
+    print(len(predict_result.candidate_matrix_poses))
+    print(predict_result.distance_matrix)
+    #import ipdb; ipdb.set_trace()
+    predict_result.visualize()
+

configs/local/inference_config.yaml

@@ -1,76 +1,76 @@
 
 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_global_only_dense
     root_dir: "experiments"
-    epoch: 1 # -1 stands for last epoch
+    epoch: 441 # -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/data/p++_dense"
+  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
+      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
+    params_name: light
 
   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: 256
     num_heads: 4
     ffn_dim: 256
     num_layers: 3
@@ -86,7 +86,8 @@ 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

@@ -22,6 +22,6 @@ runner:
   
 datasets:
     OmniObject3d:
-      root_dir: /data/hofee/nbv_rec_part2_preprocessed
-      from: 155
-      to: 165 # ..-1 means end
+      root_dir: /media/hofee/data/data/test_bottle/view 
+      from: 0
+      to: -1 # ..-1 means end

configs/local/view_generate_config.yaml

@@ -8,11 +8,11 @@ runner:
     root_dir: experiments
   generate:
     port: 5002
-    from: 600
-    to: -1 # -1 means all
-    object_dir: /media/hofee/data/data/object_meshes_part1
-    table_model_path: "/media/hofee/data/data/others/table.obj"
-    output_dir: /media/hofee/repository/data_part_1
+    from: 0
+    to: 50 # -1 means all
+    object_dir: /media/hofee/data/data/test_bottle/bottle_mesh
+    table_model_path: /media/hofee/data/data/others/table.obj
+    output_dir: /media/hofee/data/data/test_bottle/view
     binocular_vision: true
     plane_size: 10
     max_views: 512
@@ -34,7 +34,7 @@ runner:
         max_y: 0.05
         min_z: 0.01
         max_z: 0.01
-        random_rotation_ratio: 0.3
+        random_rotation_ratio: 0.0
       random_objects:
         num: 4
         cluster: 0.9

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_train_config.yaml

@@ -3,11 +3,11 @@ runner:
   general:
     seed: 0
     device: cuda
-    cuda_visible_devices: "1"
+    cuda_visible_devices: "0"
     parallel: False
     
   experiment:
-    name: debug
+    name: train_ab_global_only_with_wp_p++_strong
     root_dir: "experiments"
     use_checkpoint: False
     epoch: -1 # -1 stands for last epoch
@@ -25,87 +25,68 @@ runner:
   test:
     frequency: 3 # test frequency
     dataset_list:
-      #- OmniObject3d_test
+      - OmniObject3d_test
       - OmniObject3d_val
 
-  pipeline: nbv_reconstruction_global_pts_n_num_pipeline
+  pipeline: nbv_reconstruction_pipeline
   
 dataset:
   OmniObject3d_train:
-    root_dir: "/home/data/hofee/project/nbv_rec/data/sample_for_training_new"
+    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/sample.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
+    batch_size: 64
+    num_workers: 128
     pts_num: 8192
     load_from_preprocess: True
 
   OmniObject3d_test:
-    root_dir: "/home/data/hofee/project/nbv_rec/data/sample_for_training_new"
+    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/sample.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: 8192
     load_from_preprocess: True
 
   OmniObject3d_val:
-    root_dir: "/home/data/hofee/project/nbv_rec/data/sample_for_training_new"
+    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/sample.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: 8192
     load_from_preprocess: True
 
 
 pipeline:
-  nbv_reconstruction_local_pts_pipeline:
+  nbv_reconstruction_pipeline:
     modules:
-      pts_encoder: pointnet_encoder
+      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
 
-  nbv_reconstruction_global_pts_pipeline:
-    modules:
-      pts_encoder: pointnet_encoder
-      pose_seq_encoder: transformer_seq_encoder
-      pose_encoder: pose_encoder
-      view_finder: gf_view_finder
-    eps: 1e-5
-    global_scanned_feat: True
-
-  nbv_reconstruction_global_pts_n_num_pipeline:
-    modules:
-      pts_encoder: pointnet_encoder
-      transformer_seq_encoder: transformer_seq_encoder
-      pose_encoder: pose_encoder
-      view_finder: gf_view_finder
-      pts_num_encoder: pts_num_encoder
-    eps: 1e-5
-    global_scanned_feat: True
-
     
 module:
 
@@ -115,6 +96,10 @@ module:
     global_feat: True
     feature_transform: False
 
+  pointnet++_encoder:
+    in_dim: 3
+    params_name: strong
+
   transformer_seq_encoder:
     embed_dim: 256
     num_heads: 4
@@ -125,7 +110,7 @@ module:
   gf_view_finder:
     t_feat_dim: 128
     pose_feat_dim: 256
-    main_feat_dim: 2048
+    main_feat_dim: 5120
     regression_head: Rx_Ry_and_T
     pose_mode: rot_matrix
     per_point_feature: False

core/nbv_dataset.py

@@ -4,6 +4,7 @@ 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
@@ -34,7 +35,7 @@ class NBVReconstructionDataset(BaseDataset):
             #self.model_dir = config["model_dir"]
             self.filter_degree = config["filter_degree"]
         if self.type == namespace.Mode.TRAIN:
-            scale_ratio = 100
+            scale_ratio = 1
             self.datalist = self.datalist*scale_ratio
         if self.cache:
             expr_root = ConfigManager.get("runner", "experiment", "root_dir")
@@ -50,7 +51,7 @@ class NBVReconstructionDataset(BaseDataset):
                 scene_name_list.append(scene_name)
         return scene_name_list
 
-    def get_datalist(self):
+    def get_datalist(self, bias=False):
         datalist = []
         for scene_name in self.scene_name_list:
             seq_num = DataLoadUtil.get_label_num(self.root_dir, scene_name)
@@ -79,6 +80,8 @@ class NBVReconstructionDataset(BaseDataset):
                     for data_pair in label_data["data_pairs"]:
                         scanned_views = data_pair[0]
                         next_best_view = data_pair[1]
+                        accept_probability = scanned_views[-1][1]
+                        if accept_probability > np.random.rand():
                             datalist.append(
                                 {
                                     "scanned_views": scanned_views,
@@ -206,14 +209,9 @@ class NBVReconstructionDataset(BaseDataset):
             collate_data["combined_scanned_pts"] = torch.stack(
                 [torch.tensor(item["combined_scanned_pts"]) for item in batch]
             )
-            collate_data["scanned_pts_mask"] = torch.stack(
-                [torch.tensor(item["scanned_pts_mask"]) for item in batch]
-            )
-                
             for key in batch[0].keys():
                 if key not in [
                     "scanned_pts",
-                    "scanned_pts_mask",
                     "scanned_n_to_world_pose_9d",
                     "best_to_world_pose_9d",
                     "combined_scanned_pts",
@@ -232,9 +230,10 @@ if __name__ == "__main__":
     torch.manual_seed(seed)
     np.random.seed(seed)
     config = {
-        "root_dir": "/data/hofee/data/packed_preprocessed_data",
+        "root_dir": "/data/hofee/data/new_full_data",
+        "model_dir": "../data/scaled_object_meshes",
         "source": "nbv_reconstruction_dataset",
-        "split_file": "/data/hofee/data/OmniObject3d_train.txt",
+        "split_file": "/data/hofee/data/new_full_data_list/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

@@ -20,8 +20,8 @@ class NBVReconstructionPipeline(nn.Module):
         self.pose_encoder = ComponentFactory.create(
             namespace.Stereotype.MODULE, self.module_config["pose_encoder"]
         )
-        self.transformer_seq_encoder = ComponentFactory.create(
-            namespace.Stereotype.MODULE, self.module_config["transformer_seq_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"]
@@ -29,7 +29,6 @@ class NBVReconstructionPipeline(nn.Module):
         
 
         self.eps = float(self.config["eps"])
-        self.enable_global_scanned_feat = self.config["global_scanned_feat"]
 
     def forward(self, data):
         mode = data["mode"]
@@ -55,10 +54,7 @@ class NBVReconstructionPipeline(nn.Module):
         return perturbed_x, random_t, target_score, std
 
     def forward_train(self, data):
-        start_time = time.time()
         main_feat = self.get_main_feat(data)
-        end_time = time.time()
-        print("get_main_feat time: ", end_time - start_time)
         """ get std """
         best_to_world_pose_9d_batch = data["best_to_world_pose_9d"]
         perturbed_x, random_t, target_score, std = self.pertube_data(
@@ -79,6 +75,8 @@ class NBVReconstructionPipeline(nn.Module):
 
     def forward_test(self, data):
         main_feat = self.get_main_feat(data)
+        repeat_num = data.get("repeat_num", 1)
+        main_feat = main_feat.repeat(repeat_num, 1)
         estimated_delta_rot_9d, in_process_sample = self.view_finder.next_best_view(
             main_feat
         )
@@ -108,7 +106,7 @@ class NBVReconstructionPipeline(nn.Module):
             seq_embedding = pose_feat_seq
             embedding_list_batch.append(seq_embedding) # List(B): Tensor(S x (Dp))
         
-        seq_feat = self.transformer_seq_encoder.encode_sequence(embedding_list_batch) # Tensor(B x Ds)
+        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():

core/seq_dataset.py

@@ -2,153 +2,208 @@ 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"/home/data/hofee/project/nbv_rec/nbv_reconstruction")
+
+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_nbv_reconstruction_dataset")
-class SeqNBVReconstructionDataset(BaseDataset):
+
+@stereotype.dataset("seq_reconstruction_dataset")
+class SeqReconstructionDataset(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)
+        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.model_dir = config["model_dir"]
-        self.filter_degree = config["filter_degree"]
+        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 = []
-        for scene_name in self.scene_name_list:
-            seq_num = DataLoadUtil.get_label_num(self.root_dir, scene_name)
-            scene_max_coverage_rate = 0
+        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 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"])
+            for i in range(10,frame_len):
+                path = DataLoadUtil.get_path(self.root_dir, scene_name, i)
+                pts = DataLoadUtil.load_from_preprocessed_pts(path, "npy")
+                print(pts.shape)
+                if pts.shape[0] == 0:
+                    continue
+                else:
+                    break
+            print(i)
             datalist.append({
                 "scene_name": scene_name,
-                    "first_frame": first_frame,
-                    "max_coverage_rate": scene_max_coverage_rate,
-                    "best_seq_len": best_seq_len,
+                "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]
-        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"]
+        best_seq_len = data_item_info["best_seq_len"]
         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)
+        (
+            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_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,
+            "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,
-            "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
+    from tqdm import tqdm
+    import pickle
+    import os
+
     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,
+        "root_dir": "/media/hofee/data/data/test_bottle/view",
+        "source": "seq_reconstruction_dataset",
+        "split_file": "/media/hofee/data/data/test_bottle/test_bottle.txt",
+        "load_from_preprocess": True,
         "filter_degree": 75,
         "num_workers": 0,
-        "pts_num": 32684,
+        "pts_num": 8192,
         "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 ------+
+    
+    output_dir = "/media/hofee/data/data/test_bottle/preprocessed_dataset"
+    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": "/media/hofee/data/data/test_bottle/preprocessed_dataset",
+        "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))
+    

modules/module_lib/pointnet2_modules.py

@@ -0,0 +1,162 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from . import pointnet2_utils
+from . import pytorch_utils as pt_utils
+from typing import List
+
+
+class _PointnetSAModuleBase(nn.Module):
+
+    def __init__(self):
+        super().__init__()
+        self.npoint = None
+        self.groupers = None
+        self.mlps = None
+        self.pool_method = 'max_pool'
+
+    def forward(self, xyz: torch.Tensor, features: torch.Tensor = None, new_xyz=None) -> (torch.Tensor, torch.Tensor):
+        """
+        :param xyz: (B, N, 3) tensor of the xyz coordinates of the features
+        :param features: (B, N, C) tensor of the descriptors of the the features
+        :param new_xyz:
+        :return:
+            new_xyz: (B, npoint, 3) tensor of the new features' xyz
+            new_features: (B, npoint, \sum_k(mlps[k][-1])) tensor of the new_features descriptors
+        """
+        new_features_list = []
+
+        xyz_flipped = xyz.transpose(1, 2).contiguous()
+        if new_xyz is None:
+            new_xyz = pointnet2_utils.gather_operation(
+                xyz_flipped,
+                pointnet2_utils.furthest_point_sample(xyz, self.npoint)
+            ).transpose(1, 2).contiguous() if self.npoint is not None else None
+
+        for i in range(len(self.groupers)):
+            new_features = self.groupers[i](xyz, new_xyz, features)  # (B, C, npoint, nsample)
+
+            new_features = self.mlps[i](new_features)  # (B, mlp[-1], npoint, nsample)
+
+            if self.pool_method == 'max_pool':
+                new_features = F.max_pool2d(
+                    new_features, kernel_size=[1, new_features.size(3)]
+                )  # (B, mlp[-1], npoint, 1)
+            elif self.pool_method == 'avg_pool':
+                new_features = F.avg_pool2d(
+                    new_features, kernel_size=[1, new_features.size(3)]
+                )  # (B, mlp[-1], npoint, 1)
+            else:
+                raise NotImplementedError
+
+            new_features = new_features.squeeze(-1)  # (B, mlp[-1], npoint)
+            new_features_list.append(new_features)
+
+        return new_xyz, torch.cat(new_features_list, dim=1)
+
+
+class PointnetSAModuleMSG(_PointnetSAModuleBase):
+    """Pointnet set abstraction layer with multiscale grouping"""
+
+    def __init__(self, *, npoint: int, radii: List[float], nsamples: List[int], mlps: List[List[int]], bn: bool = True,
+                 use_xyz: bool = True, pool_method='max_pool', instance_norm=False):
+        """
+        :param npoint: int
+        :param radii: list of float, list of radii to group with
+        :param nsamples: list of int, number of samples in each ball query
+        :param mlps: list of list of int, spec of the pointnet before the global pooling for each scale
+        :param bn: whether to use batchnorm
+        :param use_xyz:
+        :param pool_method: max_pool / avg_pool
+        :param instance_norm: whether to use instance_norm
+        """
+        super().__init__()
+
+        assert len(radii) == len(nsamples) == len(mlps)
+
+        self.npoint = npoint
+        self.groupers = nn.ModuleList()
+        self.mlps = nn.ModuleList()
+        for i in range(len(radii)):
+            radius = radii[i]
+            nsample = nsamples[i]
+            self.groupers.append(
+                pointnet2_utils.QueryAndGroup(radius, nsample, use_xyz=use_xyz)
+                if npoint is not None else pointnet2_utils.GroupAll(use_xyz)
+            )
+            mlp_spec = mlps[i]
+            if use_xyz:
+                mlp_spec[0] += 3
+
+            self.mlps.append(pt_utils.SharedMLP(mlp_spec, bn=bn, instance_norm=instance_norm))
+        self.pool_method = pool_method
+
+
+class PointnetSAModule(PointnetSAModuleMSG):
+    """Pointnet set abstraction layer"""
+
+    def __init__(self, *, mlp: List[int], npoint: int = None, radius: float = None, nsample: int = None,
+                 bn: bool = True, use_xyz: bool = True, pool_method='max_pool', instance_norm=False):
+        """
+        :param mlp: list of int, spec of the pointnet before the global max_pool
+        :param npoint: int, number of features
+        :param radius: float, radius of ball
+        :param nsample: int, number of samples in the ball query
+        :param bn: whether to use batchnorm
+        :param use_xyz:
+        :param pool_method: max_pool / avg_pool
+        :param instance_norm: whether to use instance_norm
+        """
+        super().__init__(
+            mlps=[mlp], npoint=npoint, radii=[radius], nsamples=[nsample], bn=bn, use_xyz=use_xyz,
+            pool_method=pool_method, instance_norm=instance_norm
+        )
+
+
+class PointnetFPModule(nn.Module):
+    r"""Propigates the features of one set to another"""
+
+    def __init__(self, *, mlp: List[int], bn: bool = True):
+        """
+        :param mlp: list of int
+        :param bn: whether to use batchnorm
+        """
+        super().__init__()
+        self.mlp = pt_utils.SharedMLP(mlp, bn=bn)
+
+    def forward(
+            self, unknown: torch.Tensor, known: torch.Tensor, unknow_feats: torch.Tensor, known_feats: torch.Tensor
+    ) -> torch.Tensor:
+        """
+        :param unknown: (B, n, 3) tensor of the xyz positions of the unknown features
+        :param known: (B, m, 3) tensor of the xyz positions of the known features
+        :param unknow_feats: (B, C1, n) tensor of the features to be propigated to
+        :param known_feats: (B, C2, m) tensor of features to be propigated
+        :return:
+            new_features: (B, mlp[-1], n) tensor of the features of the unknown features
+        """
+        if known is not None:
+            dist, idx = pointnet2_utils.three_nn(unknown, known)
+            dist_recip = 1.0 / (dist + 1e-8)
+            norm = torch.sum(dist_recip, dim=2, keepdim=True)
+            weight = dist_recip / norm
+
+            interpolated_feats = pointnet2_utils.three_interpolate(known_feats, idx, weight)
+        else:
+            interpolated_feats = known_feats.expand(*known_feats.size()[0:2], unknown.size(1))
+
+        if unknow_feats is not None:
+            new_features = torch.cat([interpolated_feats, unknow_feats], dim=1)  # (B, C2 + C1, n)
+        else:
+            new_features = interpolated_feats
+
+        new_features = new_features.unsqueeze(-1)
+
+        new_features = self.mlp(new_features)
+
+        return new_features.squeeze(-1)
+
+
+if __name__ == "__main__":
+    pass

modules/module_lib/pointnet2_utils.py

@@ -0,0 +1,291 @@
+import torch
+from torch.autograd import Variable
+from torch.autograd import Function
+import torch.nn as nn
+from typing import Tuple
+import sys
+
+import pointnet2_cuda as pointnet2
+
+
+class FurthestPointSampling(Function):
+    @staticmethod
+    def forward(ctx, xyz: torch.Tensor, npoint: int) -> torch.Tensor:
+        """
+        Uses iterative furthest point sampling to select a set of npoint features that have the largest
+        minimum distance
+        :param ctx:
+        :param xyz: (B, N, 3) where N > npoint
+        :param npoint: int, number of features in the sampled set
+        :return:
+             output: (B, npoint) tensor containing the set
+        """
+        assert xyz.is_contiguous()
+
+        B, N, _ = xyz.size()
+        output = torch.cuda.IntTensor(B, npoint)
+        temp = torch.cuda.FloatTensor(B, N).fill_(1e10)
+
+        pointnet2.furthest_point_sampling_wrapper(B, N, npoint, xyz, temp, output)
+        return output
+
+    @staticmethod
+    def backward(xyz, a=None):
+        return None, None
+
+
+furthest_point_sample = FurthestPointSampling.apply
+
+
+class GatherOperation(Function):
+
+    @staticmethod
+    def forward(ctx, features: torch.Tensor, idx: torch.Tensor) -> torch.Tensor:
+        """
+        :param ctx:
+        :param features: (B, C, N)
+        :param idx: (B, npoint) index tensor of the features to gather
+        :return:
+            output: (B, C, npoint)
+        """
+        assert features.is_contiguous()
+        assert idx.is_contiguous()
+
+        B, npoint = idx.size()
+        _, C, N = features.size()
+        output = torch.cuda.FloatTensor(B, C, npoint)
+
+        pointnet2.gather_points_wrapper(B, C, N, npoint, features, idx, output)
+
+        ctx.for_backwards = (idx, C, N)
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_out):
+        idx, C, N = ctx.for_backwards
+        B, npoint = idx.size()
+
+        grad_features = Variable(torch.cuda.FloatTensor(B, C, N).zero_())
+        grad_out_data = grad_out.data.contiguous()
+        pointnet2.gather_points_grad_wrapper(B, C, N, npoint, grad_out_data, idx, grad_features.data)
+        return grad_features, None
+
+
+gather_operation = GatherOperation.apply
+
+
+class ThreeNN(Function):
+
+    @staticmethod
+    def forward(ctx, unknown: torch.Tensor, known: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Find the three nearest neighbors of unknown in known
+        :param ctx:
+        :param unknown: (B, N, 3)
+        :param known: (B, M, 3)
+        :return:
+            dist: (B, N, 3) l2 distance to the three nearest neighbors
+            idx: (B, N, 3) index of 3 nearest neighbors
+        """
+        assert unknown.is_contiguous()
+        assert known.is_contiguous()
+
+        B, N, _ = unknown.size()
+        m = known.size(1)
+        dist2 = torch.cuda.FloatTensor(B, N, 3)
+        idx = torch.cuda.IntTensor(B, N, 3)
+
+        pointnet2.three_nn_wrapper(B, N, m, unknown, known, dist2, idx)
+        return torch.sqrt(dist2), idx
+
+    @staticmethod
+    def backward(ctx, a=None, b=None):
+        return None, None
+
+
+three_nn = ThreeNN.apply
+
+
+class ThreeInterpolate(Function):
+
+    @staticmethod
+    def forward(ctx, features: torch.Tensor, idx: torch.Tensor, weight: torch.Tensor) -> torch.Tensor:
+        """
+        Performs weight linear interpolation on 3 features
+        :param ctx:
+        :param features: (B, C, M) Features descriptors to be interpolated from
+        :param idx: (B, n, 3) three nearest neighbors of the target features in features
+        :param weight: (B, n, 3) weights
+        :return:
+            output: (B, C, N) tensor of the interpolated features
+        """
+        assert features.is_contiguous()
+        assert idx.is_contiguous()
+        assert weight.is_contiguous()
+
+        B, c, m = features.size()
+        n = idx.size(1)
+        ctx.three_interpolate_for_backward = (idx, weight, m)
+        output = torch.cuda.FloatTensor(B, c, n)
+
+        pointnet2.three_interpolate_wrapper(B, c, m, n, features, idx, weight, output)
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_out: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        :param ctx:
+        :param grad_out: (B, C, N) tensor with gradients of outputs
+        :return:
+            grad_features: (B, C, M) tensor with gradients of features
+            None:
+            None:
+        """
+        idx, weight, m = ctx.three_interpolate_for_backward
+        B, c, n = grad_out.size()
+
+        grad_features = Variable(torch.cuda.FloatTensor(B, c, m).zero_())
+        grad_out_data = grad_out.data.contiguous()
+
+        pointnet2.three_interpolate_grad_wrapper(B, c, n, m, grad_out_data, idx, weight, grad_features.data)
+        return grad_features, None, None
+
+
+three_interpolate = ThreeInterpolate.apply
+
+
+class GroupingOperation(Function):
+
+    @staticmethod
+    def forward(ctx, features: torch.Tensor, idx: torch.Tensor) -> torch.Tensor:
+        """
+        :param ctx:
+        :param features: (B, C, N) tensor of features to group
+        :param idx: (B, npoint, nsample) tensor containing the indicies of features to group with
+        :return:
+            output: (B, C, npoint, nsample) tensor
+        """
+        assert features.is_contiguous()
+        assert idx.is_contiguous()
+
+        B, nfeatures, nsample = idx.size()
+        _, C, N = features.size()
+        output = torch.cuda.FloatTensor(B, C, nfeatures, nsample)
+
+        pointnet2.group_points_wrapper(B, C, N, nfeatures, nsample, features, idx, output)
+
+        ctx.for_backwards = (idx, N)
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_out: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        :param ctx:
+        :param grad_out: (B, C, npoint, nsample) tensor of the gradients of the output from forward
+        :return:
+            grad_features: (B, C, N) gradient of the features
+        """
+        idx, N = ctx.for_backwards
+
+        B, C, npoint, nsample = grad_out.size()
+        grad_features = Variable(torch.cuda.FloatTensor(B, C, N).zero_())
+
+        grad_out_data = grad_out.data.contiguous()
+        pointnet2.group_points_grad_wrapper(B, C, N, npoint, nsample, grad_out_data, idx, grad_features.data)
+        return grad_features, None
+
+
+grouping_operation = GroupingOperation.apply
+
+
+class BallQuery(Function):
+
+    @staticmethod
+    def forward(ctx, radius: float, nsample: int, xyz: torch.Tensor, new_xyz: torch.Tensor) -> torch.Tensor:
+        """
+        :param ctx:
+        :param radius: float, radius of the balls
+        :param nsample: int, maximum number of features in the balls
+        :param xyz: (B, N, 3) xyz coordinates of the features
+        :param new_xyz: (B, npoint, 3) centers of the ball query
+        :return:
+            idx: (B, npoint, nsample) tensor with the indicies of the features that form the query balls
+        """
+        assert new_xyz.is_contiguous()
+        assert xyz.is_contiguous()
+
+        B, N, _ = xyz.size()
+        npoint = new_xyz.size(1)
+        idx = torch.cuda.IntTensor(B, npoint, nsample).zero_()
+
+        pointnet2.ball_query_wrapper(B, N, npoint, radius, nsample, new_xyz, xyz, idx)
+        return idx
+
+    @staticmethod
+    def backward(ctx, a=None):
+        return None, None, None, None
+
+
+ball_query = BallQuery.apply
+
+
+class QueryAndGroup(nn.Module):
+    def __init__(self, radius: float, nsample: int, use_xyz: bool = True):
+        """
+        :param radius: float, radius of ball
+        :param nsample: int, maximum number of features to gather in the ball
+        :param use_xyz:
+        """
+        super().__init__()
+        self.radius, self.nsample, self.use_xyz = radius, nsample, use_xyz
+
+    def forward(self, xyz: torch.Tensor, new_xyz: torch.Tensor, features: torch.Tensor = None) -> Tuple[torch.Tensor]:
+        """
+        :param xyz: (B, N, 3) xyz coordinates of the features
+        :param new_xyz: (B, npoint, 3) centroids
+        :param features: (B, C, N) descriptors of the features
+        :return:
+            new_features: (B, 3 + C, npoint, nsample)
+        """
+        idx = ball_query(self.radius, self.nsample, xyz, new_xyz)
+        xyz_trans = xyz.transpose(1, 2).contiguous()
+        grouped_xyz = grouping_operation(xyz_trans, idx)  # (B, 3, npoint, nsample)
+        grouped_xyz -= new_xyz.transpose(1, 2).unsqueeze(-1)
+
+        if features is not None:
+            grouped_features = grouping_operation(features, idx)
+            if self.use_xyz:
+                new_features = torch.cat([grouped_xyz, grouped_features], dim=1)  # (B, C + 3, npoint, nsample)
+            else:
+                new_features = grouped_features
+        else:
+            assert self.use_xyz, "Cannot have not features and not use xyz as a feature!"
+            new_features = grouped_xyz
+
+        return new_features
+
+
+class GroupAll(nn.Module):
+    def __init__(self, use_xyz: bool = True):
+        super().__init__()
+        self.use_xyz = use_xyz
+
+    def forward(self, xyz: torch.Tensor, new_xyz: torch.Tensor, features: torch.Tensor = None):
+        """
+        :param xyz: (B, N, 3) xyz coordinates of the features
+        :param new_xyz: ignored
+        :param features: (B, C, N) descriptors of the features
+        :return:
+            new_features: (B, C + 3, 1, N)
+        """
+        grouped_xyz = xyz.transpose(1, 2).unsqueeze(2)
+        if features is not None:
+            grouped_features = features.unsqueeze(2)
+            if self.use_xyz:
+                new_features = torch.cat([grouped_xyz, grouped_features], dim=1)  # (B, 3 + C, 1, N)
+            else:
+                new_features = grouped_features
+        else:
+            new_features = grouped_xyz
+
+        return new_features

modules/module_lib/pytorch_utils.py

@@ -0,0 +1,236 @@
+import torch.nn as nn
+from typing import List, Tuple
+
+
+class SharedMLP(nn.Sequential):
+
+    def __init__(
+            self,
+            args: List[int],
+            *,
+            bn: bool = False,
+            activation=nn.ReLU(inplace=True),
+            preact: bool = False,
+            first: bool = False,
+            name: str = "",
+            instance_norm: bool = False,
+    ):
+        super().__init__()
+
+        for i in range(len(args) - 1):
+            self.add_module(
+                name + 'layer{}'.format(i),
+                Conv2d(
+                    args[i],
+                    args[i + 1],
+                    bn=(not first or not preact or (i != 0)) and bn,
+                    activation=activation
+                    if (not first or not preact or (i != 0)) else None,
+                    preact=preact,
+                    instance_norm=instance_norm
+                )
+            )
+
+
+class _ConvBase(nn.Sequential):
+
+    def __init__(
+            self,
+            in_size,
+            out_size,
+            kernel_size,
+            stride,
+            padding,
+            activation,
+            bn,
+            init,
+            conv=None,
+            batch_norm=None,
+            bias=True,
+            preact=False,
+            name="",
+            instance_norm=False,
+            instance_norm_func=None
+    ):
+        super().__init__()
+
+        bias = bias and (not bn)
+        conv_unit = conv(
+            in_size,
+            out_size,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            bias=bias
+        )
+        init(conv_unit.weight)
+        if bias:
+            nn.init.constant_(conv_unit.bias, 0)
+
+        if bn:
+            if not preact:
+                bn_unit = batch_norm(out_size)
+            else:
+                bn_unit = batch_norm(in_size)
+        if instance_norm:
+            if not preact:
+                in_unit = instance_norm_func(out_size, affine=False, track_running_stats=False)
+            else:
+                in_unit = instance_norm_func(in_size, affine=False, track_running_stats=False)
+
+        if preact:
+            if bn:
+                self.add_module(name + 'bn', bn_unit)
+
+            if activation is not None:
+                self.add_module(name + 'activation', activation)
+
+            if not bn and instance_norm:
+                self.add_module(name + 'in', in_unit)
+
+        self.add_module(name + 'conv', conv_unit)
+
+        if not preact:
+            if bn:
+                self.add_module(name + 'bn', bn_unit)
+
+            if activation is not None:
+                self.add_module(name + 'activation', activation)
+
+            if not bn and instance_norm:
+                self.add_module(name + 'in', in_unit)
+
+
+class _BNBase(nn.Sequential):
+
+    def __init__(self, in_size, batch_norm=None, name=""):
+        super().__init__()
+        self.add_module(name + "bn", batch_norm(in_size))
+
+        nn.init.constant_(self[0].weight, 1.0)
+        nn.init.constant_(self[0].bias, 0)
+
+
+class BatchNorm1d(_BNBase):
+
+    def __init__(self, in_size: int, *, name: str = ""):
+        super().__init__(in_size, batch_norm=nn.BatchNorm1d, name=name)
+
+
+class BatchNorm2d(_BNBase):
+
+    def __init__(self, in_size: int, name: str = ""):
+        super().__init__(in_size, batch_norm=nn.BatchNorm2d, name=name)
+
+
+class Conv1d(_ConvBase):
+
+    def __init__(
+            self,
+            in_size: int,
+            out_size: int,
+            *,
+            kernel_size: int = 1,
+            stride: int = 1,
+            padding: int = 0,
+            activation=nn.ReLU(inplace=True),
+            bn: bool = False,
+            init=nn.init.kaiming_normal_,
+            bias: bool = True,
+            preact: bool = False,
+            name: str = "",
+            instance_norm=False
+    ):
+        super().__init__(
+            in_size,
+            out_size,
+            kernel_size,
+            stride,
+            padding,
+            activation,
+            bn,
+            init,
+            conv=nn.Conv1d,
+            batch_norm=BatchNorm1d,
+            bias=bias,
+            preact=preact,
+            name=name,
+            instance_norm=instance_norm,
+            instance_norm_func=nn.InstanceNorm1d
+        )
+
+
+class Conv2d(_ConvBase):
+
+    def __init__(
+            self,
+            in_size: int,
+            out_size: int,
+            *,
+            kernel_size: Tuple[int, int] = (1, 1),
+            stride: Tuple[int, int] = (1, 1),
+            padding: Tuple[int, int] = (0, 0),
+            activation=nn.ReLU(inplace=True),
+            bn: bool = False,
+            init=nn.init.kaiming_normal_,
+            bias: bool = True,
+            preact: bool = False,
+            name: str = "",
+            instance_norm=False
+    ):
+        super().__init__(
+            in_size,
+            out_size,
+            kernel_size,
+            stride,
+            padding,
+            activation,
+            bn,
+            init,
+            conv=nn.Conv2d,
+            batch_norm=BatchNorm2d,
+            bias=bias,
+            preact=preact,
+            name=name,
+            instance_norm=instance_norm,
+            instance_norm_func=nn.InstanceNorm2d
+        )
+
+
+class FC(nn.Sequential):
+
+    def __init__(
+            self,
+            in_size: int,
+            out_size: int,
+            *,
+            activation=nn.ReLU(inplace=True),
+            bn: bool = False,
+            init=None,
+            preact: bool = False,
+            name: str = ""
+    ):
+        super().__init__()
+
+        fc = nn.Linear(in_size, out_size, bias=not bn)
+        if init is not None:
+            init(fc.weight)
+        if not bn:
+            nn.init.constant(fc.bias, 0)
+
+        if preact:
+            if bn:
+                self.add_module(name + 'bn', BatchNorm1d(in_size))
+
+            if activation is not None:
+                self.add_module(name + 'activation', activation)
+
+        self.add_module(name + 'fc', fc)
+
+        if not preact:
+            if bn:
+                self.add_module(name + 'bn', BatchNorm1d(out_size))
+
+            if activation is not None:
+                self.add_module(name + 'activation', activation)
+

modules/pointnet++_encoder.py

@@ -0,0 +1,148 @@
+import torch
+import torch.nn as nn
+import os
+import sys
+path = os.path.abspath(__file__)
+for i in range(2):
+    path = os.path.dirname(path)
+PROJECT_ROOT = path                 
+sys.path.append(PROJECT_ROOT)
+import PytorchBoot.stereotype as stereotype
+from modules.module_lib.pointnet2_modules import PointnetSAModuleMSG
+
+
+ClsMSG_CFG_Dense = {
+    'NPOINTS': [512, 256, 128, None],
+    'RADIUS': [[0.02, 0.04], [0.04, 0.08], [0.08, 0.16], [None, None]],
+    'NSAMPLE': [[32, 64], [16, 32], [8, 16], [None, None]],
+    'MLPS': [[[16, 16, 32], [32, 32, 64]],
+             [[64, 64, 128], [64, 96, 128]],
+             [[128, 196, 256], [128, 196, 256]],
+             [[256, 256, 512], [256, 384, 512]]],
+    'DP_RATIO': 0.5,
+}
+
+ClsMSG_CFG_Light = {
+    'NPOINTS': [512, 256, 128, None],
+    'RADIUS': [[0.02, 0.04], [0.04, 0.08], [0.08, 0.16], [None, None]],
+    'NSAMPLE': [[16, 32], [16, 32], [16, 32], [None, None]],
+    'MLPS': [[[16, 16, 32], [32, 32, 64]],
+             [[64, 64, 128], [64, 96, 128]],
+             [[128, 196, 256], [128, 196, 256]],
+             [[256, 256, 512], [256, 384, 512]]],
+    'DP_RATIO': 0.5,
+}
+
+ClsMSG_CFG_Light_2048 = {
+    'NPOINTS': [512, 256, 128, None],
+    'RADIUS': [[0.02, 0.04], [0.04, 0.08], [0.08, 0.16], [None, None]],
+    'NSAMPLE': [[16, 32], [16, 32], [16, 32], [None, None]],
+    'MLPS': [[[16, 16, 32], [32, 32, 64]],
+         [[64, 64, 128], [64, 96, 128]],
+         [[128, 196, 256], [128, 196, 256]],
+         [[256, 256, 1024], [256, 512, 1024]]],
+    'DP_RATIO': 0.5,
+}
+
+ClsMSG_CFG_Strong = {
+    'NPOINTS': [1024, 512, 256, 128, None],  # 增加采样点，获取更多细节
+    'RADIUS': [[0.02, 0.05], [0.05, 0.1], [0.1, 0.2], [0.2, 0.4], [None, None]],  # 增大感受野
+    'NSAMPLE': [[32, 64], [32, 64], [32, 64], [32, 64], [None, None]],  # 提高每层的采样点数
+    'MLPS': [[[32, 32, 64], [64, 64, 128]],  # 增强 MLP 层，增加特征提取能力
+             [[128, 128, 256], [128, 128, 256]],
+             [[256, 256, 512], [256, 384, 512]],
+             [[512, 512, 1024], [512, 768, 1024]],
+             [[1024, 1024, 2048], [1024, 1024, 2048]]],  # 增加更深的特征层
+    'DP_RATIO': 0.4,  # Dropout 比率稍微降低，以保留更多信息
+}
+
+ClsMSG_CFG_Lighter = {
+    'NPOINTS': [512, 256, 128, 64, None],
+    'RADIUS': [[0.01], [0.02], [0.04], [0.08], [None]],
+    'NSAMPLE': [[64], [32], [16], [8], [None]],
+    'MLPS': [[[32, 32, 64]],
+             [[64, 64, 128]],
+             [[128, 196, 256]],
+             [[256, 256, 512]],
+             [[512, 512, 1024]]],
+    'DP_RATIO': 0.5,
+}
+
+
+def select_params(name):
+    if name == 'light':
+        return ClsMSG_CFG_Light
+    elif name == 'lighter':
+        return ClsMSG_CFG_Lighter
+    elif name == 'dense':
+        return ClsMSG_CFG_Dense
+    elif name == 'light_2048':
+        return ClsMSG_CFG_Light_2048
+    elif name == 'strong':
+        return ClsMSG_CFG_Strong
+    else:
+        raise NotImplementedError
+
+
+def break_up_pc(pc):
+    xyz = pc[..., 0:3].contiguous()
+    features = (
+        pc[..., 3:].transpose(1, 2).contiguous()
+        if pc.size(-1) > 3 else None
+    )
+
+    return xyz, features
+
+
+@stereotype.module("pointnet++_encoder")
+class PointNet2Encoder(nn.Module):
+    def encode_points(self, pts, require_per_point_feat=False):
+        return self.forward(pts)
+
+    def __init__(self, config:dict):
+        super().__init__()
+
+        channel_in = config.get("in_dim", 3) - 3
+        params_name = config.get("params_name", "light")
+
+        self.SA_modules = nn.ModuleList()
+        selected_params = select_params(params_name)
+        for k in range(selected_params['NPOINTS'].__len__()):
+            mlps = selected_params['MLPS'][k].copy()
+            channel_out = 0
+            for idx in range(mlps.__len__()):
+                mlps[idx] = [channel_in] + mlps[idx]
+                channel_out += mlps[idx][-1]
+
+            self.SA_modules.append(
+                PointnetSAModuleMSG(
+                    npoint=selected_params['NPOINTS'][k],
+                    radii=selected_params['RADIUS'][k],
+                    nsamples=selected_params['NSAMPLE'][k],
+                    mlps=mlps,
+                    use_xyz=True,
+                    bn=True
+                )
+            )
+            channel_in = channel_out
+
+    def forward(self, point_cloud: torch.cuda.FloatTensor):
+        xyz, features = break_up_pc(point_cloud)
+
+        l_xyz, l_features = [xyz], [features]
+        for i in range(len(self.SA_modules)):
+            li_xyz, li_features = self.SA_modules[i](l_xyz[i], l_features[i])
+            l_xyz.append(li_xyz)
+            l_features.append(li_features)
+        return l_features[-1].squeeze(-1)
+
+
+if __name__ == '__main__':
+    seed = 100
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+    net = PointNet2Encoder(config={"in_dim": 3, "params_name": "strong"}).cuda()
+    pts = torch.randn(2, 2444, 3).cuda()
+    print(torch.mean(pts, dim=1))
+    pre = net.encode_points(pts)
+    print(pre.shape)

preprocess/preprocessor.py

@@ -164,10 +164,10 @@ def save_scene_data(root, scene, scene_idx=0, scene_total=1,file_type="txt"):
 
 if __name__ == "__main__":
     #root = "/media/hofee/repository/new_data_with_normal"
-    root = r"H:\AI\Datasets\nbv_rec_part2"
+    root = r"/media/hofee/data/data/test_bottle/view"
     scene_list = os.listdir(root)
     from_idx = 0 # 1000
-    to_idx = 600 # 1500
+    to_idx = len(scene_list) # 1500
 
 
     cnt = 0

runners/inference_server.py

@@ -12,8 +12,9 @@ from PytorchBoot.runners.runner import Runner
 from PytorchBoot.utils import Log
 
 from utils.pts import PtsUtil
+from beans.predict_result import PredictResult
 
-@stereotype.runner("inferencer")
+@stereotype.runner("inferencer_server")
 class InferencerServer(Runner):
     def __init__(self, config_path):
         super().__init__(config_path)
@@ -24,40 +25,45 @@ class InferencerServer(Runner):
         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("nbv_evaluator")
+        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)
-        fps_downsampled_combined_scanned_pts, fps_idx = PtsUtil.fps_downsample_point_cloud(
-            combined_scanned_views_pts, self.pts_num, require_idx=True
+        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
         )
-        combined_scanned_views_pts_mask = np.zeros(len(scanned_pts), dtype=np.uint8)
-        start_idx = 0
-        for i in range(len(scanned_pts)):
-            end_idx = start_idx + len(scanned_pts[i])
-            combined_scanned_views_pts_mask[start_idx:end_idx] = i
-            start_idx = end_idx
 
-        fps_downsampled_combined_scanned_pts_mask = combined_scanned_views_pts_mask[fps_idx]
-
-        input_data["scanned_pts_mask"] = np.asarray(fps_downsampled_combined_scanned_pts_mask, dtype=np.uint8)
+        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):
 
-        estimated_delta_rot_9d = output_data["pred_pose_9d"]
+        pred_pose_9d = output_data["pred_pose_9d"]
+        pred_pose_9d = np.asarray(PredictResult(pred_pose_9d.cpu().numpy(), None, cluster_params=dict(eps=0.25, min_samples=3)).candidate_9d_poses, dtype=np.float32)
         result = {
-            "estimated_delta_rot_9d": estimated_delta_rot_9d.tolist()
+            "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))
     
@@ -65,7 +71,8 @@ class InferencerServer(Runner):
         def inference():
             data = request.json
             input_data = self.get_input_data(data)
-            output_data = self.pipeline.forward_test(input_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)
         

runners/inferencer.py

@@ -4,6 +4,7 @@ from utils.render import RenderUtil
 from utils.pose import PoseUtil
 from utils.pts import PtsUtil
 from utils.reconstruction import ReconstructionUtil
+from beans.predict_result import PredictResult
 
 import torch
 from tqdm import tqdm
@@ -19,14 +20,19 @@ 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)
@@ -34,6 +40,11 @@ class Inferencer(Runner):
             
         ''' 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 '''
@@ -65,128 +76,182 @@ 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_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
                         
-                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)
+                        self.save_inference_result(test_set_name, data["scene_name"], output)
+                    except Exception as e:
+                        Log.error(f"Error in scene {scene_name}, {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["combined_scanned_pts"] = torch.tensor(data["first_scanned_pts"][0], dtype=torch.float32).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_data["combined_scanned_pts"] = data["combined_scanned_pts"]
-        input_pts_N = input_data["scanned_pts"][0].shape[1]
+        input_pts_N = input_data["combined_scanned_pts"].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)
+        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]
-        last_pred_cr = self.compute_coverage_rate(scanned_view_pts, None, down_sampled_model_pts, threshold=voxel_threshold)
-
+        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)
-
+            # # save pred_pose_9d ------
+            # root = "/media/hofee/data/project/python/nbv_reconstruction/nbv_reconstruction/temp_output_result"
+            # scene_dir = os.path.join(root, scene_name)
+            # if not os.path.exists(scene_dir):
+            #     os.makedirs(scene_dir)
+            # pred_9d_path = os.path.join(scene_dir,f"pred_pose_9d_{len(pred_cr_seq)}.npy")
+            # pts_path = os.path.join(scene_dir,f"combined_scanned_pts_{len(pred_cr_seq)}.txt")
+            # np_combined_scanned_pts = input_data["combined_scanned_pts"][0].cpu().numpy()
+            # np.save(pred_9d_path, pred_pose_9d.cpu().numpy())
+            # np.savetxt(pts_path, np_combined_scanned_pts)
+            # # ----- ----- -----
+            predict_result = PredictResult(pred_pose_9d.cpu().numpy(), input_pts=input_data["combined_scanned_pts"][0].cpu().numpy(), cluster_params=dict(eps=0.25, min_samples=3))
+            # -----------------------
+            # import ipdb; ipdb.set_trace()
+            # predict_result.visualize()
+            # -----------------------
+            pred_pose_9d_candidates = predict_result.candidate_9d_poses
+            for pred_pose_9d in pred_pose_9d_candidates:
+                #import ipdb; ipdb.set_trace()
+                pred_pose_9d = torch.tensor(pred_pose_9d, dtype=torch.float32).to(self.device).unsqueeze(0)
                 pred_pose[:3,:3] = PoseUtil.rotation_6d_to_matrix_tensor_batch(pred_pose_9d[:,:6])[0]
                 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, _ = 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_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]
+                scanned_view_pts.append(new_target_pts)
                 
-            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)]
                 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)
+                
+                combined_scanned_pts = np.vstack(scanned_view_pts)
+                voxel_downsampled_combined_scanned_pts_np = PtsUtil.voxel_downsample_point_cloud(combined_scanned_pts, voxel_threshold)
                 random_downsampled_combined_scanned_pts_np = PtsUtil.random_downsample_point_cloud(voxel_downsampled_combined_scanned_pts_np, input_pts_N)
                 input_data["combined_scanned_pts"] = torch.tensor(random_downsampled_combined_scanned_pts_np, dtype=torch.float32).unsqueeze(0).to(self.device)
 
                 
                 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
+                break
             
 
-        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
 
@@ -199,6 +264,13 @@ class Inferencer(Runner):
         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 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 save_inference_result(self, dataset_name, scene_name, output):
         dataset_dir = os.path.join(self.output_dir, dataset_name)
@@ -206,7 +278,7 @@ 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)
     
 

utils/data_load.py

@@ -24,8 +24,6 @@ class DataLoadUtil:
         for channel in float_channels:
             channel_data = exr_file.channel(channel)
             img_data.append(np.frombuffer(channel_data, dtype=np.float16).reshape((height, width)))
-
-        # 将各通道组合成一个 (height, width, 3) 的 RGB 图像
         img = np.stack(img_data, axis=-1)
         return img
 

utils/pts.py

@@ -17,6 +17,17 @@ class PtsUtil:
             unique_voxels = np.unique(voxel_indices, axis=0, return_inverse=True)
             return unique_voxels[0]*voxel_size
         
+    @staticmethod
+    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):
         if point_cloud.shape[0] == 0:

utils/reconstruction.py

@@ -32,13 +32,15 @@ class ReconstructionUtil:
         
     
     @staticmethod
-    def check_overlap(new_point_cloud, combined_point_cloud, overlap_area_threshold=25, voxel_size=0.01):
+    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 < voxel_size*2)
+        overlapping_points_num = np.sum(distances < voxel_size*2)
         cm = 0.01
         voxel_size_cm = voxel_size / cm
-        overlap_area = overlapping_points * 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
 
 

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 +85,52 @@ 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)
+            #import ipdb; ipdb.set_trace()
             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)
+            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
 
-            ''' 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)
+            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

@@ -7,6 +7,7 @@ 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
+from utils.pose import PoseUtil
 
 class visualizeUtil:
     
@@ -34,6 +35,21 @@ class visualizeUtil:
         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 get_cam_pose_and_cam_axis(cam_pose, is_6d_pose):
+        if is_6d_pose:
+            matrix_cam_pose = np.eye(4)
+            matrix_cam_pose[:3,:3] = PoseUtil.rotation_6d_to_matrix_numpy(cam_pose[:6])
+            matrix_cam_pose[:3, 3] = cam_pose[6:]
+        else:
+            matrix_cam_pose = cam_pose
+        cam_pos = matrix_cam_pose[:3, 3]
+        cam_axis = matrix_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)
+        return cam_pos, sample_points
+    
     @staticmethod
     def save_all_combined_pts(root, scene, output_dir):
         length = DataLoadUtil.get_scene_seq_length(root, scene)