Reproduce “Attentive Illumination Decomposition Model for Multi-Illuminant White Balancing” 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
Dongyoung Kim, Jinwoo Kim, Junsang Yu, Seon Joo Kim
The Attentive Illumination Decomposition (AID) model offers a deep learning framework that explicitly decomposes complex multi-illuminant scenes into individual light source chromaticities and their spatial influence maps. This approach yields state-of-the-art white balancing performance and enables controllable color adjustments by adhering to a physically grounded light mixing model.
Key insights
- Decomposing illumination into individual sources and their spatial influence, rather than direct mixed illumination prediction, provides a more physically grounded and interpretable approach to multi-illuminant white balancing.
- Slot attention, previously successful in object-centric learning, can be effectively adapted to disentangle fundamental image properties like illumination, leading to structured and interpretable representations.
- The proposed centroid-matching loss is crucial for guiding slots to specialize in specific color ranges, enabling accurate decomposition and significantly improving the model's ability to identify distinct illuminants.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.