Swin Transformer for Color Constancy.

In order to fully extract the illumination information, this work proposes a color constancy algorithm based on Swin Transformer. The algorithm calculates the self attention by taking full account of the relationship between pixels. This work improves the ability of the algorithm to extract the feature information of the illumination and estimates the field illumination more accurately. On the reprocessed Color Checker datasets, the algorithm in this work outperforms the convolutional neural network in many indicators, proving the accuracy and robustness of the algorithm.