VisionCoach: Reinforcing Grounded Video Reasoning via Visual-Perception Prompting

Benchmark Model Rank Results
video-question-answering-on-videommmuVisionCoach-7B#10Accuracy: 54.4
audio-visual-question-answering-on-worldsenseVisionCoach-7B#16Accuracy: 43.8
video-question-answering-on-video-mmmuVisionCoach-7B#26Accuracy: 54.4
zero-shot-video-question-answer-on-video-mme-w-o-subsVisionCoach-7B#40Accuracy (%): 63.3
moment-retrieval-on-charades-staVisionCoach-7B#75mIoU: 42.7