Wanglong Lu

Wanglong Lu

Nasdaq

EchoSR: Efficient Context Harnessing for Lightweight Image Super-Resolution

Published in Information Fusion, 104471, 2026

Hanli Zhao, Binhao Wang, Shihao Zhao, Tao Wang, Kaihao Zhang, Wanglong Lu

EchoSR architecture

Brief description:

EchoSR is an efficient context-harnessing framework for lightweight image super-resolution. It decouples feature learning into local, multi-scale, and global modeling stages, then uses cross-scale overlapping fusion to integrate fine-grained details and high-level structures.

Highlights:

  • Context-harnessing block with local aggregation, multi-scale receptive field expansion, and global context prior.
  • Cross-scale overlapping fusion for coherent information propagation across spatial scales.
  • Designed for the reconstruction-quality and computational-efficiency trade-off in resource-constrained SR.
  • Reported to outperform lightweight SR baselines across multiple benchmarks while achieving about 2x faster speed.

EchoSR efficiency comparison

[paper] [arxiv] [pdf] [html] [github]

Recommended citation:

@article{zhao2026echosr,
  title={EchoSR: Efficient Context Harnessing for Lightweight Image Super-Resolution},
  author={Zhao, Hanli and Wang, Binhao and Zhao, Shihao and Wang, Tao and Zhang, Kaihao and Lu, Wanglong},
  journal={Information Fusion},
  pages={104471},
  year={2026},
  doi={10.1016/j.inffus.2026.104471}
}