Using Neural Factorization of Shape and Reflectance for Ball Detection

Visually detecting a well-defined object like a soccer ball should be a simple problem. Under good lighting conditions this problem can be claimed to be solved, but unfortunately, the lighting conditions are not always optimal. In those circumstances, it is valuable to have a shape and reflectance model of the ball, to be able to predict how its appearance changes if the lighting changes. This is a prerequisite for playing soccer outside, with direct sunlight and clouds. The predicted appearance will be used to fine-tune an existing ball detection algorithm, based on the classic Yolo algorithm.