Watch Before You Answer: Learning from Visually Grounded Post-Training

Benchmark Model Rank Results
video-question-answering-on-videommmuVidGround (64 frames)#12Accuracy: 49.4
video-question-answering-on-mmvuVidGround (32 frames)#21Accuracy: 65.8
zero-shot-video-question-answer-on-video-mme-w-o-subsVidGround (64 frames)#38Accuracy (%): 63.4