Blind All In One Image Restoration

Blind Unified Image Restoration aims to remove various degradations from input images without prior knowledge of the type or severity of degradation. This task is evaluated under two settings: three-degradation and five-degradation. The three-degradation setting includes rain removal, dehazing, and denoising; the five-degradation setting adds defogging and low-light enhancement on top of these. The training dataset covers Rain200L, RESIDE, WED, BSD400, GoPro, and LoLv1, while the evaluation datasets use Rain100L, SOTS (outdoor), BSD68, GoPro, and LoLv1. Performance metrics are measured by reporting the average PSNR and SSIM across all evaluation datasets, highlighting the model's overall capability in handling diverse degradations. This task emphasizes the versatility and robustness of image restoration techniques.