Reproduce “Fast Fourier Color Constancy” 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
Jonathan T. Barron, Yun-Ta Tsai
Google researchers developed Fast Fourier Color Constancy (FFCC), an algorithm that processes log-chroma histograms with Fast Fourier Transforms to achieve state-of-the-art accuracy in illuminant estimation. The method operates at over 700 frames per second on mobile devices, providing probabilistic outputs suitable for robust real-time auto white balance.
Key insights
- Framing color constancy as a 2D localization task on a small, toroidal log-chroma plane enables highly efficient processing via FFTs, despite requiring specialized 'illuminant de-aliasing' techniques.
- Differentiable fitting of bivariate von Mises distributions allows for end-to-end training of models that predict both illuminant estimates and their uncertainties from toroidal probability maps.
- A novel Fourier-domain regularization and preconditioning technique significantly accelerates training (20x faster) and improves the quality of learned models by optimizing periodic parameters.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.