VideoLLaMA 3: Frontier Multimodal Foundation Models for Image and Video Understanding

Benchmark Model Rank Results
video-question-answering-on-next-qaVideoLLaMA3(7B)#3Accuracy: 84.5
video-question-answering-on-perception-test-testVideoLLaMA3 7B#3Accuracy: 72.8
visual-question-answering-vqa-on-infographicvqaVideoLLaMA3 7B#9ANLS: 78.9
visual-question-answering-vqa-on-mmeVideoLLaMA3 7B#11Score: 2102
visual-question-answering-vqa-on-mmeVideoLLaMA3 2B#18Score: 1901
visual-question-answering-on-docvqa-testVideoLLaMA3 7B#12ANLS: 0.949
visual-question-answering-on-docvqa-testVideoLLaMA3 2B#30ANLS: 0.919
video-question-answering-on-tempcompass-overallVideoLLaMA3 7B#13Accuracy: 68.1
zero-shot-video-question-answer-on-egoschema-fullsetVideoLLaMA3 7B#19Accuracy: 63.3
video-question-answering-on-video-mme-with-subtitlesVideoLLaMA3 7B (w/ subs)#25Accuracy: 70.3
video-question-answering-on-mlvuVideoLLaMA3 7B#28Accuracy: 73.0
video-question-answering-on-mmvuVideoLLaMA3 7B#31Accuracy: 44.1
video-question-answering-on-mvbenchVideoLLaMA3 7B#31Avg.: 69.7
chart-question-answering-on-chartqaVideoLLaMA3 7B#331:1 Accuracy: 86.3
chart-question-answering-on-chartqaVideoLLaMA3 2B#691:1 Accuracy: 79.8
zero-shot-video-question-answer-on-video-mme-w-o-subsVideoLLaMA3 7B#36Accuracy (%): 66.2
visual-question-answering-vqa-on-realworldqaVideoLLaMA3 7B#37Accuracy: 72.7
visual-question-answering-vqa-on-realworldqaVideoLLaMA3 2B#65Accuracy: 67.3
video-question-answering-on-lvbenchVideoLLaMA3 7B#43Accuracy: 45.3
optical-character-recognition-ocr-on-ocrbenchVideoLLaMA3 7B#44Score: 828
visual-question-answering-vqa-on-blinkVideoLLaMA3 2B#47Accuracy: 44.2
multimodal-reasoning-on-mathvisionVideoLLaMA3 7B#54Accuracy: 26.2
multimodal-reasoning-on-mathvisionVideoLLaMA3 2B#62Accuracy: 15.5
multimodal-reasoning-on-mmmu-proVideoLLaMA3 7B#58Accuracy: 33.6
multimodal-reasoning-on-mmmu-proVideoLLaMA3 2B#59Accuracy: 28.6
multimodal-reasoning-on-mathvistaVideoLLaMA3 7B#75Accuracy: 67.1
multimodal-reasoning-on-mathvistaVideoLLaMA3 2B#93Accuracy: 59.2
multimodal-reasoning-on-mmmuVideoLLaMA3 7B#100Accuracy: 48.8
multimodal-reasoning-on-mmmuVideoLLaMA3 2B#111Accuracy: 45.3