Reproduce “Perception-Inspired Color Space Design for Photo White Balance Editing” 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
Yang Cheng, Ziteng Cui, Shenghan Su, Lin Gu, Zenghui Zhang
Researchers from Shanghai Jiao Tong University, The University of Tokyo, and Tohoku University introduce a perception-inspired learnable HSI color space and a dual-branch attention network for white balance correction in sRGB images. Their method achieves state-of-the-art performance on in-distribution datasets and features an exceptionally compact model size of 6.40 MB.
Key insights
- Adapting the color space itself to the specific task of white balance correction, rather than relying on fixed transformations, enables better handling of ISP non-linearities.
- A learnable luminance axis and adaptive non-linear mapping functions in LHSI allow for more accurate disentanglement of image features, aiding color constancy.
- Co-designing a network architecture, DCLAN, to process these disentangled luminance and chromatic channels through MambaVision and cross-attention optimizes performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.