Problem with scripts/vis.py about multi_part_assembly

[EdoardoBonanni]

My setup is to use pointnet coding, RGL-net as model, task assembly of type geometric and artifact as dataset.

[Wuziyi616]

Can you put your config file here, so that I can run it to check what's wrong?

EDIT: or can you provide more information? E.g. the full error message indicating which line is wrong.

[crtie]

I meet the same question, here's my full error message
Traceback (most recent call last):
File "scripts/vis.py", line 134, in
visualize(cfg)
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/autograd/grad_mode.py", line 28, in decorate_context
return func(*args, **kwargs)
File "scripts/vis.py", line 36, in visualize
out_dict = model(batch) # trans/rot: [B, P, 3/4/(3, 3)]
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
return forward_call(*input, **kwargs)
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/parallel/data_parallel.py", line 169, in forward
return self.gather(outputs, self.output_device)
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/parallel/data_parallel.py", line 181, in gather
return gather(outputs, output_device, dim=self.dim)
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/parallel/scatter_gather.py", line 78, in gather
res = gather_map(outputs)
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/parallel/scatter_gather.py", line 69, in gather_map
return type(out)((k, gather_map([d[k] for d in outputs]))
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/parallel/scatter_gather.py", line 69, in
return type(out)((k, gather_map([d[k] for d in outputs]))
File "/home/vipuser/miniconda3/envs/PA_tcr/lib/python3.8/site-packages/torch/nn/parallel/scatter_gather.py", line 73, in gather_map
return type(out)(map(gather_map, zip(*outputs)))
File "/home/vipuser/multi_part_assembly/multi_part_assembly/utils/rotation.py", line 119, in init
self._check_valid()
File "/home/vipuser/multi_part_assembly/multi_part_assembly/utils/rotation.py", line 139, in _check_valid
assert isinstance(self._rot, torch.Tensor), 'rotation must be a tensor'
AssertionError: rotation must be a tensor

[Wuziyi616]

Thanks for the info. Which model are you testing? Is this problem associated with one model or all of them?

[crtie]

I trained dgl and LSTM following your usage.md and try to vis their results, but both model met this problem.

[Wuziyi616]

That's weird. I try the vis.py but I can pass that line (and trigger a bug in the line after loss_dict, _ = model.module._calc_loss(out_dict, batch)) Have you pull the newest code from master branch? Or are you using dev branch?

[crtie]

yeah，I pulled the newest code from master branch just yesterday. And it goes well when I run train.py and test.py

[crtie]

okay! I already fixed this bug by cancel the DataParallel wrpper. I know this may decrease efficienfy, but it's enough for me. Thank for your prompt reply.
Here is my changed vis.py

[Wuziyi616]

Ah, I think I understand where the problem is. Yes, the current code won't work with multi-gpu. I'll fix it. Thanks for providing insights here!

[Wuziyi616]

Fixed in the latest commit