Reproduce “Learning to Sample the Most Useful Training Patches from Images” 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
Shuyang Sun, Liang Chen, Gregory Slabaugh, Philip Torr
PatchNet improves demosaicing by learning during training which image patches to emphasize, rather than relying on random crops or manual hard-example mining. It weights patch losses while training a restoration network and is not used at inference. On MIT Moire, with training on Vimeo-90k, PatchNet-large reaches 33.73 dB PSNR, 2.35 dB above DeepJDD; gains are specific to this comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.