OpenGVLab / Vision-RWKV

Vision-RWKV: Efficient and Scalable Visual Perception with RWKV-Like Architectures
https://arxiv.org/abs/2403.02308
Apache License 2.0
288 stars 11 forks source link

Extension of WKV from 1D to 2D. #24

Open Yaziwel opened 2 weeks ago

Yaziwel commented 2 weeks ago

It is an impressive extension of WKV from unidirectional to bidirectional. However, the current WKV is still limited to 1D sequences. I believe the relative bias |t-i|/T can be extended to (|x-i|/H + |y-j|/W), where x and y represent two directions in an image. It appears that this approach can also be transformed into a recurrent form with linear complexity.