mit-han-lab / spvnas

[ECCV 2020] Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution
http://spvnas.mit.edu/
MIT License
587 stars 109 forks source link

Fix bug in `visualize.py` and add Open3D-based visualization #87

Closed suyunzzz closed 2 years ago

suyunzzz commented 2 years ago

Fix a bug in visualize.py and add Open3D-based visualization.

zhijian-liu commented 2 years ago

Thanks for the contribution! The newly added visualizer is not used right now. Is open3d an alternative to mlab? If so, could you please add an option so that we can choose the visualization backend.

suyunzzz commented 2 years ago

hello,zhijian, I am a git newbie, i dont know whether "push -f" is right, if there are some mistake please tell me,:)🤣

QuasarsZ commented 2 years ago

Hello, @suyunzzz I've downloaded the file which you updated yesterday. And I try to run the visualize code by default command:

python visualize.py

But I encounter some error about:

File "visualize.py", line 360, in outputs = model(inputs) ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 84])

The parse argument is:

Model: SemanticKITTI_val_SPVNAS@65GMACs (default) Velodyne_dir: sample_data(default) visualize_backend: open3d(default)

Do you know how to solve this problem? Hope you can give me some advise. Thanks a lot!! 谢谢!

zhijian-liu commented 2 years ago

Hi @suyunzzz, thanks for the updates. Could you please remove init and all .pyc files? Also, the sample data is now stored under assets. Please remove sample_data as well. Thank you!

zhijian-liu commented 2 years ago

Btw, could you also reformat the code according to pre-commit (see https://github.com/mit-han-lab/spvnas/runs/6327544860?check_suite_focus=true for more details)? Thanks!

zhijian-liu commented 2 years ago

Hello, @suyunzzz I've downloaded the file which you updated yesterday. And I try to run the visualize code by default command:

python visualize.py

But I encounter some error about:

File "visualize.py", line 360, in outputs = model(inputs) ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 84])

The parse argument is:

Model: SemanticKITTI_val_SPVNAS@65GMACs (default) Velodyne_dir: sample_data(default) visualize_backend: open3d(default)

Do you know how to solve this problem? Hope you can give me some advise. Thanks a lot!! 谢谢!

@QuasarsZ, could you please double-check the input size: inputs.F.shape?

QuasarsZ commented 2 years ago

Hello, @zhijian-liu. Thanks for your response.

I run the visualize code with:

python visualize.py --velodyne-dir ./dataset/semantic-kitti/00/velodyne/

And the input size is:

torch.Size([91852, 4])

Is my velodyne-dir set incorrectly?

suyunzzz commented 2 years ago

Hello, @zhijian-liu. Thanks for your response.

I run the visualize code with:

python visualize.py --velodyne-dir ./dataset/semantic-kitti/00/velodyne/

And the input size is:

torch.Size([91852, 4])

Is my velodyne-dir set incorrectly?

Does it work if you use a original visualize.py?

suyunzzz commented 2 years ago

Btw, could you also reformat the code according to pre-commit (see https://github.com/mit-han-lab/spvnas/runs/6327544860?check_suite_focus=true for more details)? Thanks! hello, zhijian, I am not sure that is need format code manually? or automatic?🤣

QuasarsZ commented 2 years ago

Hi, @suyunzzz. Thank's for your reply.

Run original visualize.py will get another error. (#90)

Then, I only modify this:

inds, labels, inverse_map = sparsequantize(pc,
feat_,
voxel_size,
return_index=True, return_inverse=True)

to

coords_, inds, inverse_map = sparsequantize(pc, return_index=True, return_inverse=True)

And I will get error message about:

File "visualize.py", line 360, in outputs = model(inputs) ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 84])

CCInc commented 2 years ago

@suyunzzz To use precommit,

pip install pre-commit
pre-commit install
pre-commit run -a

It should reformat the code and then format the code before every commit.

suyunzzz commented 2 years ago

@suyunzzz To use precommit,

pip install pre-commit
pre-commit install
pre-commit run -a

It should reformat the code and then format the code before every commit.

thanks, it seems work😄