Position aware circular convolution

🚀 The feature

Recently, we propose a new basic operation position aware global circular convolution (ParC). Differing from previous convolution operations, the proposed ParC has global receptive field. Experimental results show that ParC uniformly
improves performance of various typical models.

This work has been accpted by ECCV 2022, we hope the proposed ParC can be used by other researchers. Please refer to refer to https://arxiv.org/abs/2203.03952 to find more details.

Motivation, pitch

We are suggested to reach out to torchvision when we post this issue in pytorch.

https://github.com/pytorch/pytorch/issues/80932#issuecomment-1178268557

Alternatives

We will prepare a CUDA optimized version ParC following tutorial in https://pytorch.org/tutorials/advanced/cpp_extension.html.

We also want to propose a pull request to merge our code into torchvision. Will such a pull request be accpted ? If it is acceptable, is there anyone can provide us a demo of proposing this kind of pull request?

Additional context

No response

The issue names no repository files, tests, or entry points. Start with the linked ECCV 2022 paper and the PyTorch C++ extension tutorial, then confirm with torchvision maintainers whether this operation belongs here and what API and tests are required. Done is a maintainer-approved scope or a concrete PR path.