PointClouds2Volumes got '_LinAlgError: linalg.inv:' error when setting grid_size = [440 500 1]

[Uzukidd]

Hi, I try seting grid_size = [440 500 1] and get an error, it seems that bugs exist when performing 2D voxelization.

_LinAlgError                              Traceback (most recent call last)
Cell In[15], [line 16](vscode-notebook-cell:?execution_count=15&line=16)
      [8](vscode-notebook-cell:?execution_count=15&line=8) initial_volumes = Volumes(
      [9](vscode-notebook-cell:?execution_count=15&line=9)     features = torch.zeros(1, 3, grid_size[0], grid_size[1], grid_size[2]),
     [10](vscode-notebook-cell:?execution_count=15&line=10)     densities = torch.zeros(1, 1, grid_size[0], grid_size[1], grid_size[2]),
     [11](vscode-notebook-cell:?execution_count=15&line=11)     volume_translation = [0.0, 0.0, 0.0],
     [12](vscode-notebook-cell:?execution_count=15&line=12)     voxel_size = VOXEL_SIZE.tolist(),  
     [13](vscode-notebook-cell:?execution_count=15&line=13) )
     [14](vscode-notebook-cell:?execution_count=15&line=14) # add the pointcloud to the 'initial_volumes' buffer using
     [15](vscode-notebook-cell:?execution_count=15&line=15) # trilinear splatting
---> [16](vscode-notebook-cell:?execution_count=15&line=16) updated_volumes = add_pointclouds_to_volumes(
     [17](vscode-notebook-cell:?execution_count=15&line=17)     pointclouds=pointclouds,
     [18](vscode-notebook-cell:?execution_count=15&line=18)     initial_volumes=initial_volumes,
     [19](vscode-notebook-cell:?execution_count=15&line=19)     mode="nearest",
     [20](vscode-notebook-cell:?execution_count=15&line=20) )

File [d:\anaconda3\envs\env_beampy_torch\lib\site-packages\pytorch3d\ops\points_to_volumes.py:282](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:282), in add_pointclouds_to_volumes(pointclouds, initial_volumes, mode, min_weight, rescale_features, _python)
    [279](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:279) mask = (mask[None, :] < n_per_pcl[:, None]).type_as(mask)
    [281](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:281) # convert to the coord frame of the volume
--> [282](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:282) pcl_3d_local = initial_volumes.world_to_local_coords(pcl_3d)
    [284](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:284) features_new, densities_new = add_points_features_to_volume_densities_features(
    [285](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:285)     points_3d=pcl_3d_local,
    [286](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/ops/points_to_volumes.py:286)     points_features=pcl_feats,
   (...)
...
    [294](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/transforms/transform3d.py:294)     Return the inverse of self._matrix.
    [295](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/transforms/transform3d.py:295)     """
--> [296](file:///D:/anaconda3/envs/env_beampy_torch/lib/site-packages/pytorch3d/transforms/transform3d.py:296)     return torch.inverse(self._matrix)

_LinAlgError: linalg.inv: (Batch element 0): The diagonal element 1 is zero, the inversion could not be completed because the input matrix is singular.