d-li14/involution
A novel neural operator that inverts the spatial-specific channel-general property of convolution, designed for visual recognition tasks.

This repository provides the official implementation of involution, a general-purpose neural primitive published at CVPR 2021. Involution serves as an alternative to standard convolution, designed to be more efficient and effective for deep learning models. The operator bridges concepts from convolution and self-attention, and can be applied across a range of vision tasks including image classification, object detection, instance segmentation, and semantic segmentation. Built on the OpenMMLab framework with pretrained models.