search

mit-han-lab / torchsparse

[MICRO'23, MLSys'22] TorchSparse: Efficient Training and Inference Framework for Sparse Convolution on GPUs.
https://torchsparse.mit.edu
MIT License
1.16k stars 132 forks source link

Fix batch_size #235

Closed dailing57 closed 10 months ago

dailing57 commented 10 months ago

TorchSparse v2.1 changed the coordinate order from [x, y, z, batch] to [batch, x, y, z]. The batch_size acquisition method needs to be modified

ys-2020 commented 10 months ago

Hi @dailing57 , thank you very much for the catch! We really appreciate your effort. We have already fixed the bug in the wheels on our pypi server. However, since the source code for v2.1.0 is still under internal check, we haven't release it yet. The code you have changed is actually v2.0. Therefore, we may not be able to merge this PR now.

Once again, thank you for your interest and effort in TorchSparse.

Best regards, Shang