V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

Benchmark Model Rank Results
video-question-answering-on-tempcompass-mcqV-JEPA 2 ViT-g (1B, 384px) + Llama 3.1 (8B)#1Accuracy: 76.9
video-question-answering-on-tomatoV-JEPA 2 ViT-g (1B, 384px) + Llama 3.1 (8B)#1Accuracy: 40.3
video-question-answering-on-tvbenchV-JEPA 2 ViT-g 8B#2Average Accuracy: 60.6