Reproduce “RAW-Adapter: Adapting Pre-trained Visual Model to Camera RAW 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

Ziteng Cui, Tatsuya Harada

RAW-Adapter adapts camera-sensor RAW images for visual tasks while reusing sRGB-pretrained models. It uses a compact, task-oriented ISP-like input pipeline and connects intermediate ISP-stage information to adapters in the downstream network. On synthetic dark PASCAL RAW detection, RetinaNet with ResNet-50 reaches 86.6 mAP, versus 83.6 for Dirty-Pixel; normal-light scores tie, and results are limited to tested settings.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.