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

non-negative coords only? #107

Closed thomas-enxuli closed 1 year ago

thomas-enxuli commented 1 year ago

Hi @zhijian-liu , thanks for the amazing work! Quick question regarding the coordinates here. I see that in https://github.com/mit-han-lab/spvnas/blob/master/core/datasets/semantic_kitti.py#L198

pc_ -= pc_.min(0, keepdims=1)

The coordinates are made to be non-negative. Is there a constraint that voxel indices for sparse tensor or point tensor to be non-negative? Thanks

zhijian-liu commented 1 year ago

It's a common practice to normalize the coordinates to be non-negative. This is because some sparse convolution libraries do not support negative indices.