DisParQ

New method DisParQ enables interpretable vision models without labels

Researchers have developed DisParQ, a novel self-supervised method for creating interpretable vision foundation models. This approach learns spatially grounded, discrete concept representations without requiring class l…