Reproduce “Recolour What Matters: Region-Aware Colour Editing via Token-Level Diffusion” 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
Yuqi Yang, Dongliang Chang, Yijia Ling, Ruoyi Du, Zhanyu Ma
The ColourCrafter framework, from Beijing University of Posts and Telecommunications, introduces a token-level diffusion model for fine-grained, region-aware colour editing that unifies semantic localization from text with continuous chromatic control from explicit RGB references. It achieved superior colour accuracy and perceptual quality in quantitative metrics and user studies compared to existing methods, consistently reproducing target hues within specified regions while preserving structural and textural fidelity.
Key insights
- Unifying semantic localization from text guidance with continuous chromatic precision from explicit RGB references at a token level within a diffusion model enables accurate, region-aware colour propagation.
- Applying a perceptually uniform Lab-space constraint, masked to specific editing regions, effectively mitigates colour bias and enhances pixel-level chromatic fidelity by supervising only the relevant areas.
- Decoupling luminance and chrominance in the loss function helps preserve realistic shading, brightness, and structural integrity of the image while precisely adjusting hue in edited regions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.