In this post we discuss SwAV (Swapping Assignments between multiple Views of the same image) method from the paper "Unsupervised Learning of Visual Features by Contrasting Cluster Assignments" by M. Caron et al. For those interested in coding, several code repositories about SwAV algorithm are on GitHub; if in doubt, take a look at the repo mentioned in the paper. Supervised learning works with labeled training data; for example, in supervised image classification algorithms a cat photo needs to be annotated and labeled as "cat". Self-supervised learning aims at obtaining features without using manual annotations. We will consider, in paticular, visual features.
Jan-24-2022, 10:45:55 GMT