LLaVA-OneVision: Easy Visual Task Transfer

Benchmark Model Rank Results
3d-question-answering-3d-qa-on-scanqa-test-wLLaVA-NeXT-Video#14Exact Match: 18.7BLEU-4: 9.8ROUGE: 27.8METEOR: 9.1CIDEr: 46.2
3d-question-answering-3d-qa-on-sqa3dLLaVA-NeXT-Video#8Exact Match: 34.2
on-implicitqaLLaVA-OneVision - 7B#3Average Accuracy: 43.4Macro Average Accuracy: 46.4
temporal-relation-extraction-on-vinogroundLLaVA-OneVision-Qwen2-72B#2Text Score: 48.4Video Score: 35.2Group Score: 21.8
temporal-relation-extraction-on-vinogroundLLaVA-OneVision-Qwen2-7B#3Text Score: 41.6Video Score: 29.4Group Score: 14.6
video-question-answering-on-next-qaLLaVA-OV(72B)#12Accuracy: 80.2
video-question-answering-on-next-qaLLaVA-OV(7B)#14Accuracy: 79.4
video-question-answering-on-ovbenchLLaVA-OneVision (7B)#4AVG: 49.5
visual-question-answering-on-mm-vetLLaVA-OneVision-72B#19GPT-4 score: 63.7
visual-question-answering-on-mm-vetLLaVA-OneVision-7B#34GPT-4 score: 57.5
visual-question-answering-on-mm-vetLLaVA-OneVision-0.5B#152GPT-4 score: 29.1
visual-question-answering-on-v-benchLLaVA-OneVision7B#4Accuracy: 74.46
visual-question-answering-vqa-on-vlm2-benchLLaVA-OneVision-7B#6GC-mat: 16.60GC-trk: 13.70OC-cpr: 47.22OC-cnt: 56.17
zero-shot-video-question-answer-on-vnbenchLLaVA-OneVision-72B#3Accuracy: 58.7
zero-shot-video-question-answer-on-vnbenchLLaVA-OneVision-7B#4Accuracy: 51.8