UNSUPERVISED SAM FOR AUTOMATIC IMAGE SEGMENTATION

Summary from the TLDR AI newsletter

Unsupervised SAM (UnSAM) is a new model for whole-image segmentation that eliminates the need for human annotations. Using a divide-and-conquer approach, UnSAM identifies hierarchical structures in visual scenes by combining top-down and bottom-up clustering methods.