RuntimeError occurs when running PartA2 on torch 1.2

Hello,

I found that the following error will occur when running PartA2 on torch 1.2.

RuntimeError: range.second - range.first == t.size() INTERNAL ASSERT FAILED at /pytorch/torch/csrc/autograd/generated/Functions.cpp:55, please report a bug to PyTorch. inconsistent range for TensorList output

Since the requirements in INSTALL.md says that the code was tested on PyTorch 1.1, I changed the PyTorch version and it worked.

The issue here is quite similar. Is function torch.nonzero() the reason?

Thank you.

[gardenbaby]

Yeah. It still works when I test PartA2 model but not when training, while PVRCNN works in both situations. Here is the problem.

RuntimeError: range.second - range.first == t.size() INTERNAL ASSERT FAILED at /opt/conda/conda-
bld/pytorch_1565272271120/work/torch/csrc/autograd/generated/Functions.cpp:55, 
please report a bug to PyTorch. inconsistent range for TensorList output

Same here. The error only occurs when training PartA2 in PCDet v0.2. I changed the PyTorch version from 1.2 to 1.1 and it worked.

[sshaoshuai]

It is reasonable that fg_inds is empty during the beginning of the training, and we do have functions to process this situation as follows: https://github.com/open-mmlab/OpenPCDet/blob/master/pcdet/models/roi_heads/target_assigner/proposal_target_layer.py#L149-L154