torch-points3d icon indicating copy to clipboard operation
torch-points3d copied to clipboard

scannet dataset not working for object_detection task

Open fcakyon opened this issue 3 years ago • 0 comments

steps to reproduce:

inside vscode remote container using the .devcontainer config:

poetry run python train.py task=object_detection models=object_detection/votenet2 model_name=VoteNetMink data=object_detection/scannet-sparse

error:

Exception has occurred: TypeError       (note: full exception trace is shown but execution is paused at: main)
sequence item 1: expected str instance, NoneType found
  File "/workspaces/pytorch-pointcloud-detector/torch_points3d/datasets/dataset_factory.py", line 23, in get_dataset_class
    dataset_module = ".".join(["torch_points3d.datasets", task, module])
  File "/workspaces/pytorch-pointcloud-detector/torch_points3d/datasets/dataset_factory.py", line 45, in instantiate_dataset
    dataset_cls = get_dataset_class(dataset_config)
  File "/workspaces/pytorch-pointcloud-detector/torch_points3d/trainer.py", line 91, in _initialize_trainer
    self._dataset: BaseDataset = instantiate_dataset(self._cfg.data)
  File "/workspaces/pytorch-pointcloud-detector/torch_points3d/trainer.py", line 44, in __init__
    self._initialize_trainer()
  File "/workspaces/pytorch-pointcloud-detector/train.py", line 13, in main (Current frame)
    trainer = Trainer(cfg)
  File "/workspaces/pytorch-pointcloud-detector/train.py", line 22, in <module>
    main()

reason:

There is no task field in the dataset_config when loading scannet dataset.

It may be broken after the omegaconf2 update https://github.com/nicolas-chaulet/torch-points3d/pull/604

@CCInc @nicolas-chaulet

fcakyon avatar Oct 17 '21 06:10 fcakyon