Brief

Researchers have introduced CanonCGT, a novel two-stage framework for reference-based color grading. This method utilizes a canonical pivot, an intermediate style-neutral representation, to ensure stable color mapping and preserve scene structure. CanonCGT aims to overcome the instability and unnatural results of existing techniques by first canonicalizing input images and then applying the reference style. The framework employs a dual-phase training scheme that combines supervised learning with self-supervised refinement, demonstrating superior photorealism and tonal consistency. AI