#vision representation

Meta AI introduces Web-SSL, a family of large-scale visual self-supervised models trained without language supervision. These models achieve competitive results on multimodal benchmarks, challenging the need for language in vision learning.