DINOv2: Learning Robust Visual Features

Main Problem Addressed The Introduction argues that computer vision still lacks a true “foundation model” equivalent to large language models: a system that produces universally useful visual feature