"padding" operation for the model

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

1.19k stars 139 forks source link

"padding" operation for the model #225

Closed christy0621 closed 1 year ago

christy0621 commented 1 year ago

Hi guys,

I'm new to the field and I'm trying to reimplement Cylinder3D based on torchsparse package. However, I don't find the operation "padding" based on the supported function.

Could I get some help to implement that? Some snip code would be super helpful.

Thanks in advance.

Cheers

zhijian-liu commented 1 year ago

I don't think we have circular padding implemented. You probably need to manually pad the input before sending into our convolution.

christy0621 commented 1 year ago

Cool, thanks for the reply anyway.