Comment by heyitsguay

As a computer vision guy I'm sad JEPA didn't end up more effective. Makes perfect sense conceptually, would have easily transferred to video, but other self-supervised methods just seem to beat it!