UniRestorer++ scales image restoration with degradation experts
Original title: UniRestorer++: Scalable Universal Image Restoration with Multi-Granularity Degradation Modeling
The paper reports roughly log-linear PSNR gains and a 50-category CDD-50 benchmark.
Researchers propose UniRestorer++, a universal image restoration framework that grows capacity through multi-granularity degradation modeling and group-specific experts. Routing uses degradation estimates to pick a root-to-leaf path and granularity estimates to choose a level, with dense maps for spatially uneven corruption. They introduce CDD-50, covering 50 degradation categories from single to penta-degradation, and report stronger scaling on out-of-distribution and real-world data than vanilla mixture-of-experts scaling.
- The model claims state-of-the-art results and zero-shot plus few-shot transfer.
- Code is at github.com/mrluin/UniRestorer_plusplus.
- Vanilla deeper and wider backbones are described as saturating.
From the original article
1 min read at huggingface.co
Universal image restoration aims to handle diverse degradations within a single model, but a reliable scaling strategy remains underexplored. Increasing backbone depth and width can lead to performance saturation, while degradation-aware specialization depends on accurate degradation estimation and is vulnerable to errors on out-of-distribution and real-world inputs.
By Jingbo Lin, Zhilu Zhang, Fan Li · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.