DINOv3

2025-08-13

AI Summary

This work addresses the construction of general-purpose vision foundation models, aiming to eliminate reliance on manual annotations while enabling cross-domain (natural and aerial imagery) representation learning and scalable training over large-scale data and model sizes. To mitigate dense feature map degradation during long-horizon training, we propose a Gram anchoring mechanism; additionally, a multi-scale post-adaptation strategy is introduced to enhance flexibility in resolution handling, parameter scaling, and text–image alignment. Our method builds upon a self-supervised contrastive learning framework, integrating large-scale heterogeneous data preprocessing, feature map regularization, and lightweight post-processing techniques. The resulting model achieves state-of-the-art performance across classification, detection, and segmentation tasks—without task-specific fine-tuning—outperforming domain-specialized SOTA methods. Moreover, its dense feature representations exhibit substantially higher quality than those of existing self-supervised and weakly supervised foundation models.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Deep Neural Architectures and Foundation Models
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Self-supervised learning holds the promise of eliminating the need for manual data annotation, enabling models to scale effortlessly to massive datasets and larger architectures. By not being tailored to specific tasks or domains, this training paradigm has the potential to learn visual representations from diverse sources, ranging from natural to aerial images -- using a single algorithm. This technical report introduces DINOv3, a major milestone toward realizing this vision by leveraging simple yet effective strategies. First, we leverage the benefit of scaling both dataset and model size by careful data preparation, design, and optimization. Second, we introduce a new method called Gram anchoring, which effectively addresses the known yet unsolved issue of dense feature maps degrading during long training schedules. Finally, we apply post-hoc strategies that further enhance our models' flexibility with respect to resolution, model size, and alignment with text. As a result, we present a versatile vision foundation model that outperforms the specialized state of the art across a broad range of settings, without fine-tuning. DINOv3 produces high-quality dense features that achieve outstanding performance on various vision tasks, significantly surpassing previous self- and weakly-supervised foundation models. We also share the DINOv3 suite of vision models, designed to advance the state of the art on a wide spectrum of tasks and data by providing scalable solutions for diverse resource constraints and deployment scenarios.

Problem

• Eliminate manual data annotation with self-supervised learning
• Address dense feature map degradation during long training
• Enhance model flexibility in resolution and text alignment

Innovation

• Scaling dataset and model size effectively
• Introducing Gram anchoring for stable training
• Enhancing model flexibility post-training