Let's Discuss DINO-v1 Research
ykilcher · x · 2026-07-19
The post invites everyone to discuss DINO-v1 tonight at 6:00 PM UTC. The attached image is the paper "Emerging Properties in Self-Supervised Vision Transformers", whose core conclusion is that without using supervised labels, the last-layer self-attention of ViT can automatically learn class-specific features and present effects applicable to unsupervised object segmentation.
More from Research
- Researchers release 44B synthetic tokens for higher-quality pretraining data — vanstriendaniel · 2026-07-21
- ReViV reconstructs egocentric 4D viewer-and-scene dynamics from one monocular video — ethz-vlg · 2026-07-21
- A hand-worked batch norm example shows exactly what gets normalized and why — ProfTomYeh · 2026-07-21
- Google Research says diffusion creativity is a byproduct of smooth score learning — dl_weekly · 2026-07-21
- Open reproduction of Meta’s REWIRE data pipeline cuts the cost to about $11 — vanstriendaniel · 2026-07-21
- Nat Lambert finishes an RLHF book with a 10-hour course and training code — natolambert · 2026-07-21