Skip to content

v1.4.0
Compare
Choose a tag to compare
@zhijian-liu zhijian-liu released this 25 Jun 01:03
· 39 commits to master since this release
v1.4.0
74099d1
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

New Features

  • Supported mixed-precision training and inference with torch.cuda.amp (#69, #75).
  • Added generalized sparse convolution (#77).
  • Added group normalization (#63).

API Changes

  • sparse_{collate,quantize} now needs to be imported from torchsparse.utils.{collate,quantize}.
  • sparse_quantize now takes in coords, voxel_size (defaults to 1.0), return_index (defaults to False) and return_inverse (defaults to False) as input, and returns the quantized coords as well as indices and inverse_indices (if requested).
  • transpose is renamed as transposed for torchsparse.nn.Conv3d and torchsparse.nn.functional.conv3d.
  • KernelRegion is removed and replaced with get_kernel_offsets (in torchsparse.nn.utils).
Assets 2
Loading