Meta’s Vision Foundation Model DINOv3 Returns as the Champion: Self-Supervision Surpasses Weak Supervision for the First Time, Commercially Available and Open Source

Meta has released and open-sourced DINOv3, a state-of-the-art vision foundation model based on self-supervised learning. For the first time, DINOv3 has outperformed weakly supervised models across multiple vision tasks. Trained on 1.7 billion images with 7 billion parameters, its innovative self-supervised learning approach eliminates reliance on labeled data and reduces training costs.