Self-Supervised Tracking via Video Colorization

Google researchers managed to learn object tracking without labeled data by training a neural network to colorize grayscale videos. By starting off with a single colored frame, the model learned to color all subsequent frames, which now allows tracking objects by initially coloring them. It can follow multiple objects, track through occlusions, and remain robust over deformations without requiring any labeled training data.


Want to receive more content like this in your inbox?