V-JEPA models trained only on feature prediction from 2 million public videos achieve 81.9% on Kinetics-400, 72.2% on Something-Something-v2, and 77.9% on ImageNet-1K using frozen ViT-H/16 backbones.
preprint arXiv:2006.06882 , year=
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.CV 2representative citing papers
Adding register tokens to Vision Transformers eliminates high-norm background artifacts and raises state-of-the-art performance on dense visual prediction tasks.
citing papers explorer
-
Revisiting Feature Prediction for Learning Visual Representations from Video
V-JEPA models trained only on feature prediction from 2 million public videos achieve 81.9% on Kinetics-400, 72.2% on Something-Something-v2, and 77.9% on ImageNet-1K using frozen ViT-H/16 backbones.
-
Vision Transformers Need Registers
Adding register tokens to Vision Transformers eliminates high-norm background artifacts and raises state-of-the-art performance on dense visual prediction tasks.