Cube++ is a novel dataset collected for illumination estimation problem. It has 4890 raw 18-megapixel images, each containing a SpyderCube color target in their scenes, manually labelled categories, and ground truth illumination chromaticities.

[CVPR2020] A Multi-Hypothesis Approach to Color Constancy

An official TensorFlow implementation of “CLCC: Contrastive Learning for Color Constancy” accepted at CVPR 2021.

Not a serious implementation of Deep white balance in Tensorflow. Aimed for personal learning.

Source code and dataset for the paper titled "Colour alignment for relative colour constancy via non-standard references"

Implementation of the method described in the paper "Quasi-unsupervised color constancy" - CVPR 2019

Reference code for the paper Auto White-Balance Correction for Mixed-Illuminant Scenes.

Reference code for the paper: Deep White-Balance Editing (CVPR 2020). Our method is a deep learning multi-task framework for white-balance editing.

Reference code for the paper "Cross-Camera Convolutional Color Constancy" (ICCV 2021)

A suite of tests to assess attention faithfulness for explainability

A Convolutional Framework for Color Constancy [IEEE TNNLS 2024]