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

torchsparse 2.1.0 missing PointTensor #213

Closed BenicioEnriquez closed 1 year ago

BenicioEnriquez commented 1 year ago

Is there an existing issue for this?

Current Behavior

I just attempted to use torchsparse with NeuralRecon, and found that I am unable to proceed because PointTensor is completely missing from the package. There is no documentation as to why it was removed or where it's now located. There isn't even source code for 2.1.0, only 2.0.0. I am unable to install an older version because there are only wheels for 2.1.0 online.

Expected Behavior

Either documentation on whereabouts or deprecation of PointTensor, replacement of PointTensor back into torchsparse (if it's an error), or access to older wheel files.

Environment

- GCC: 9.4.0
- NVCC: 11.7
- PyTorch: 1.13.1
- PyTorch CUDA: 11.7
- TorchSparse: 2.1.0+torch113cu117

Anything else?

No response

kentang-mit commented 1 year ago

Hi @TetraEngine,

This is the expected behavior. We deprecated PointTensor from torchsparse because it is only used in the SPVNAS project. We instead provided a better implementation of this class in spvnas, which can be accessed here.

Best, Haotian

440981 commented 11 months ago

你好@TetraEngine,

这是预期的行为。我们弃用PointTensor了 torchsparse,因为它仅在 SPVNAS 项目中使用。相反,我们在 spvnas 中提供了此类的更好实现,可以在此处访问。

最好的, 昊天

你好@TetraEngine,

这是预期的行为。我们弃用PointTensor了 torchsparse,因为它仅在 SPVNAS 项目中使用。相反,我们在 spvnas 中提供了此类的更好实现,可以在此处访问。

最好的, 昊天

How can I introduce or use alternatives