LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding

Benchmark Model Rank Results
zero-shot-video-question-answer-on-video-mmeLongVU (7B)#10Accuracy (%): 60.6
zero-shot-video-question-answer-on-egoschema-fullsetLongVU (7B)#11Accuracy: 67.6
video-question-answering-on-mvbenchLongVU (7B)#47Avg.: 66.9
video-question-answering-on-mlvuLongVU#51Accuracy: 65.4