Vision Transformer with Deformable Attention

Deformable Attention (wip)

Implementation of Deformable Attention in Pytorch from this paper, which appears to be better than what was proposed in DETR. This repository may also explore 1d and 3d cases, as well as fix the relative positional bias to extrapolate better (SwinV2 style)

Citation

@misc{xia2022vision,
    title   = {Vision Transformer with Deformable Attention}, 
    author  = {Zhuofan Xia and Xuran Pan and Shiji Song and Li Erran Li and Gao Huang},
    year    = {2022},
    eprint  = {2201.00520},
    archivePrefix = {arXiv},
    primaryClass = {cs.CV}
}

GitHub

View Github