I am attempting to use the PointPillarNet trainable file pointpillars_trainable.tlt found here: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/models/pointpillarnet
To the best of my understanding, it is supposed to be used as a pre-trained model for the base PointPillars version part of the tao toolkit. Is this correct based off of this section of the documentation?:
I see a field for pretrained_model_path in the model portion of the pointpillars.yaml file and filled it in with this:
model:
name: PointPillar
pretrained_model_path: pointpillars_trainable.tlt
The rest of the yaml is the same as the default. When I try to run the train command, I get this error:
INFO: Starting PointPillars training
INFO: Start logging
INFO: CUDA_VISIBLE_DEVICES=ALL
INFO: Database filter by min points Car: 14357 => 13461
INFO: Database filter by min points Pedestrian: 2207 => 2161
INFO: Database filter by min points Cyclist: 734 => 700
INFO: Loading point cloud dataset
INFO: Total samples for point cloud dataset: 3712
/usr/local/lib/python3.10/dist-packages/torch/functional.py:507: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument. (Triggered internally at /opt/pytorch/pytorch/aten/src/ATen/native/TensorShape.cpp:3549.)
return _VF.meshgrid(tensors, **kwargs) # type: ignore[attr-defined]
WARNING: ‘decrypt_stream’ is deprecated, to be removed in ‘0.7’. Please use ‘eff.codec.decrypt_stream()’ instead.
INFO: unregistered extension code 18968
Traceback (most recent call last):
File “/usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/pointcloud/pointpillars/scripts/train.py”, line 203, in
raise e
File “/usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/pointcloud/pointpillars/scripts/train.py”, line 187, in
main()
File “/usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/pointcloud/pointpillars/scripts/train.py”, line 128, in main
model, opt_state, start_epoch, it = build_model_and_optimizer(
File “/usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/pointcloud/pointpillars/pcdet/models/init.py”, line 126, in build_model_and_optimizer
pretrained_model = load_checkpoint(pretrained_model_path, key, to_cpu)[0]
File “/usr/local/lib/python3.10/dist-packages/nvidia_tao_pytorch/pointcloud/pointpillars/pcdet/models/init.py”, line 61, in load_checkpoint
loaded_model = torch.load(temp_name, map_location=loc_type)
File “/usr/local/lib/python3.10/dist-packages/torch/serialization.py”, line 1038, in load
return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
File “/usr/local/lib/python3.10/dist-packages/torch/serialization.py”, line 1256, in _legacy_load
magic_number = pickle_module.load(f, **pickle_load_args)
ValueError: unregistered extension code 18968
Here is my yaml file saved as a text file:
pointpillars.txt (5.6 KB)
2 posts - 2 participants