Self-Supervised Models Outperform Supervised in Art Classification
A new study from arXiv systematically compares supervised and self-supervised feature extractors for artwork classification and retrieval, focusing on paintings. Using DINO family and CLIP models, the researchers found that self-supervised backbones consistently improve classification performance. The work also explores real-world applications like VR museum navigation. The paper is available under Computer Vision and Pattern Recognition.
Key facts
- Study compares supervised vs self-supervised backbones for art classification
- Focus on paintings using DINO family and CLIP models
- Self-supervised backbones consistently improve classification performance
- Applications include VR museum navigation
- Published on arXiv under Computer Vision and Pattern Recognition
- Submission history and code/data available
- arXivLabs framework mentioned for community collaboration
- Paper ID: 2605.18974
Entities
Institutions
- arXiv
- arXivLabs