Reproduce “Wonder 4: A Mixture-of-Experts Recipe for Image Quality Enhancement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yunan Wu, Bin Sun, Partha Acharjee, Xiaoyu Wang
Wonder 4 tackles mixed-defect photo restoration by routing each image patch to five of 13 learned correction experts in each model block. It uses a minimum-use floor to raise selection chances for underused experts, rather than enforcing equal use. On DIV2K wild ×4 upscaling, it reports 18% lower LPIPS and 16% lower DISTS than Qwen-Image-Edit, but does not lead PSNR.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.