The official implementation of "RSRWKV: A Linear-Complexity 2D Attention Mechanism for Efficient Remote Sensing Vision Task".
2025/09/23: We release the code and models of RSRWKV.
RSRWKV is a novel adaptation of the RWKV architecture for high-resolution remote sensing analysis.
It introduces a Linear-Complexity 2D Attention Mechanism through the 2D-WKV scanning strategy, enabling efficient isotropic context aggregation. The framework integrates:
- 2D-WKV: bridges sequential processing with spatial reasoning.
- MVC-Shift Module: enhances multiscale receptive field coverage.
- Efficient Channel Attention (ECA): improves cross-channel interaction and semantic saliency.
Experiments on NWPU RESISC45, VHR-10 v2, SSDD, and GLHWater demonstrate superior performance over CNN and Transformer baselines in classification, detection, and segmentation.
All model weights and logs are available at Baidu Drive.
If this work is helpful for your research, please consider citing the following BibTeX entry.
@article{li2025rsrwkv,
title={RSRWKV: A Linear-Complexity 2D Attention Mechanism for Efficient Remote Sensing Vision Task},
author={Li, Chunshan and Wang, Rong and Yang, Xiaofei and Chu, Dianhui},
journal={arXiv preprint arXiv:2503.20382},
year={2025}
}