Computer Vision News - May 2020

2 Summary Co puter Vision Challenge 30 Color constancy is the way that we perceive the color of objects irrespective of the illumination of a scene. The goal of computational color constancy is to reproduce this property. One way to achieve this is to overcome the problem of illumination chromaticity estimation. Egor tells us he is trying to trick his camera. “It’s good enough,” he says, “but if you get closer to it, the colors on the screen will change. The picture gets warmer. This means that the white balance algorithm for this camera is pretty bad. The challenge is developing these algorithms.” He adds that color correction stable to the content of the scene is important. “You should try to guess the answer about the chromaticity of the source , and it should be good enough for industry,” he explains. This year’s challenge will have two major differences. “First of all, it will have a bigger data set,” Alex tells us. While the 2019 challenge included 2,000 images, which was a much larger data set than before, industry "Color constancy is the way that we perceive the color of objects irrespective of the illumination of a scene." Alex Savchik Simon Karpenko Egor Ershov