VideoMind: A Chain-of-LoRA Agent for Long Video Reasoning

Benchmark Model Rank Results
zero-shot-video-question-answer-on-next-gqaVideoMind(7B)#2Acc@GQA: 28.2
zero-shot-video-question-answer-on-next-gqaVideoMind (2B)#5Acc@GQA: 25.2
natural-language-moment-retrieval-on-tacosVideoMind 7B (extra data)#15R@1,IoU=0.3: 49.5R@1,IoU=0.5: 36.2R@1,IoU=0.7: 21.4mIoU: 34.4
moment-retrieval-on-charades-staVideoMind 7B (zero-shot) (extra data)#37R@1 IoU=0.5: 59.1R@1 IoU=0.7: 31.2R@1 IoU=0.3: 73.5mIoU: 50.2