Learnable adaptive bilateral filter for improved generalization in Single Image Super-Resolution

Wenhao Guo, Peng Lu, Xujun Peng, Zhaoran Zhao

Abstract: Highlights•Introduced an adaptive bilateral filter-based operator to reduce image distribution gaps, boosting SR generalization.•Proposed HMA-RL framework for weakly supervised parameter search of the bilateral filter, solving LR-clean pair issue.•Developed a simple multi-agent reward strategy to enhance pixel-level collaboration and preprocessing performance.•HAM-RL-based SR network achieves top results on real-world datasets with fewer samples and reduced resources.