DINOv3: Self-supervised learning for vision at unprecedented scale

DINOv3 revolutionizes computer vision by achieving state-of-the-art performance across diverse tasks without human-labeled data. This self-supervised learning model, scaled to 7B parameters and trained on 1.7B images, produces superior high-resolution visual features that outperform specialized solutions in object detection and segmentation while enabling lightweight adapters for resource-efficient deployment in healthcare, environmental monitoring, and satellite imagery analysis.