Reinforced Attention Learning

Benchmark Model Rank Results
video-question-answering-on-lvbenchRAL-zero–Accuracy: 45.9
video-question-answering-on-tempcompass-overallOn-Policy Distillation + Attention Distillation–Accuracy: 70.3
video-question-answering-on-video-mmmuRAL-zero–Accuracy: 49.2
visual-question-answering-vqa-on-blinkRAL–Accuracy: 57.2
zero-shot-video-question-answer-on-video-mme-w-o-subsRAL-zero–Accuracy (%): 65.1