Small Vision-Language Models are Smart Compressors for Long Video Understanding

Benchmark Model Rank Results
zero-shot-video-question-answer-on-video-mmeTempo (4K Budget)#4Accuracy (%): 67.8
zero-shot-video-question-answer-on-video-mmeTempo (8K Budget)#5Accuracy (%): 67.7
video-question-answering-on-mlvuTempo#22Accuracy: 75.6
zero-shot-video-question-answer-on-video-mme-w-o-subsTempo#30Accuracy (%): 67.8