Self-supervised learning for vision

Level 6: Transformers and multimodal vision

Coming soon

This page is being written. See the full roadmap for the whole syllabus.

What this chapter will cover

• Why learn without labels
• pretext tasks
• contrastive learning (SimCLR, MoCo)
• non-contrastive methods (BYOL, DINO)
• masked image modeling (MAE)
• linear probing and k-NN evaluation
• DINOv2 features