upd

configs/server/server_train_config.yaml

@@ -3,11 +3,11 @@ runner:
   general:
     seed: 0
     device: cuda
-    cuda_visible_devices: "0"
+    cuda_visible_devices: "1"
     parallel: False
     
   experiment:
-    name: overfit_ab_global_only
+    name: overfit_ab_local_only
     root_dir: "experiments"
     use_checkpoint: False
     epoch: -1 # -1 stands for last epoch
@@ -39,7 +39,7 @@ dataset:
     type: train
     cache: True
     ratio: 1
-    batch_size: 80
+    batch_size: 32
     num_workers: 16
     pts_num: 8192
     load_from_preprocess: True
@@ -55,7 +55,7 @@ dataset:
     eval_list:
       - pose_diff
     ratio: 1
-    batch_size: 80
+    batch_size: 32
     num_workers: 12
     pts_num: 8192
     load_from_preprocess: True
@@ -71,7 +71,7 @@ dataset:
     eval_list:
       - pose_diff
     ratio: 1
-    batch_size: 80
+    batch_size: 32
     num_workers: 12
     pts_num: 8192
     load_from_preprocess: True
@@ -92,16 +92,16 @@ module:
 
   pointnet_encoder:
     in_dim: 3
-    out_dim: 1024
+    out_dim: 512
     global_feat: True
     feature_transform: False
 
   transformer_seq_encoder:
-    embed_dim: 256
+    embed_dim: 768
     num_heads: 4
     ffn_dim: 256
     num_layers: 3
-    output_dim: 1024
+    output_dim: 2048
 
   gf_view_finder:
     t_feat_dim: 128

core/nbv_dataset.py

@@ -165,13 +165,8 @@ class NBVReconstructionDataset(BaseDataset):
             [best_to_world_6d, best_to_world_trans], axis=0
         )
 
-        combined_scanned_views_pts = np.concatenate(scanned_views_pts, axis=0)
-        voxel_downsampled_combined_scanned_pts_np = PtsUtil.voxel_downsample_point_cloud(combined_scanned_views_pts, 0.002)
-        random_downsampled_combined_scanned_pts_np = PtsUtil.random_downsample_point_cloud(voxel_downsampled_combined_scanned_pts_np, self.pts_num)
-        
         data_item = {
             "scanned_pts": np.asarray(scanned_views_pts, dtype=np.float32), # Ndarray(S x Nv x 3)
-            "combined_scanned_pts": np.asarray(random_downsampled_combined_scanned_pts_np, dtype=np.float32), # Ndarray(N x 3)
             "scanned_coverage_rate": scanned_coverages_rate, # List(S): Float, range(0, 1)
             "scanned_n_to_world_pose_9d": np.asarray(scanned_n_to_world_pose, dtype=np.float32), # Ndarray(S x 9)
             "best_coverage_rate": nbv_coverage_rate, # Float, range(0, 1)
@@ -203,15 +198,13 @@ class NBVReconstructionDataset(BaseDataset):
             collate_data["best_to_world_pose_9d"] = torch.stack(
                 [torch.tensor(item["best_to_world_pose_9d"]) for item in batch]
             )
-            collate_data["combined_scanned_pts"] = torch.stack(
+                
-                [torch.tensor(item["combined_scanned_pts"]) for item in batch]
-            )
             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",
                 ]:
                     collate_data[key] = [item[key] for item in batch]
             return collate_data

core/pipeline.py

@@ -29,6 +29,7 @@ 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"]
@@ -91,24 +92,23 @@ class NBVReconstructionPipeline(nn.Module):
         scanned_n_to_world_pose_9d_batch = data[
             "scanned_n_to_world_pose_9d"
         ]  # List(B): Tensor(S x 9)
-
+        scanned_pts_batch = data[
+            "scanned_pts"
+        ]
         device = next(self.parameters()).device
 
         embedding_list_batch = []
 
-        combined_scanned_pts_batch = data["combined_scanned_pts"]  # Tensor(B x N x 3)
+        for scanned_n_to_world_pose_9d, scanned_pts in zip(scanned_n_to_world_pose_9d_batch, scanned_pts_batch):
-        global_scanned_feat = self.pts_encoder.encode_points(
-            combined_scanned_pts_batch, require_per_point_feat=False
-        )  # global_scanned_feat: Tensor(B x Dg)
-
-        for scanned_n_to_world_pose_9d in scanned_n_to_world_pose_9d_batch:
             scanned_n_to_world_pose_9d = scanned_n_to_world_pose_9d.to(device)  # Tensor(S x 9)
+            scanned_pts = scanned_pts.to(device)  # Tensor(S x N x 3)
             pose_feat_seq = self.pose_encoder.encode_pose(scanned_n_to_world_pose_9d)  # Tensor(S x Dp) 
-            seq_embedding = pose_feat_seq
+            pts_feat_seq = self.pts_encoder.encode_points(scanned_pts, require_per_point_feat=False)  # Tensor(S x Dl)
-            embedding_list_batch.append(seq_embedding) # List(B): Tensor(S x (Dp))
+            seq_embedding = torch.cat([pose_feat_seq, pts_feat_seq], dim=-1) # Tensor(S x (Dp+Dl))
+            embedding_list_batch.append(seq_embedding) # List(B): Tensor(S x (Dp+Dl))
         
         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))
+        main_feat = seq_feat # Tensor(B x Ds)
 
         if torch.isnan(main_feat).any():
             Log.error("nan in main_feat", True)