Add RoI ops for Torchvision #1290
Comments
github-merge-queue bot
pushed a commit
that referenced
this issue
Jan 21, 2025
Adds SYCL kernels for `roi_pool`, `ps_roi_pool`, and `ps_roi_align` ops, and registers them in torchvision. Code is adapted from existing CUDA kernels, with minimal changes, similar to the existing `nms` and `roi_align` ops. Part of #1290 ### Testing For testing, I copied the comprehensive ops [test ](https://github.com/pytorch/vision/blob/main/test/test_ops.py) from torchvision repo. I removed the DefornConv test (support to come in a follow-up PR) and autocast xpu tests, since autocast registration in Pytorch would need to be updated.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
🚀 The feature, motivation and pitch
Currently, XPU supports
torchvision::nmsandtorchvision::roi_align, but additional ops in the torchvision repo are not supported. My proposal is to add SYCL kernels for the following ops:ps_roi_alignroi_poolps_roi_pooldeform_conv2dAlternatives
No response
Additional context
No response
The text was updated successfully, but these errors were encountered: