Persistent Object Narratives for Token-Efficient Video Language Models

Benchmark Model Rank Results
video-question-answering-on-activitynet-qaSlotNarrative–Accuracy: 50.3Confidence score: 3.5
zeroshot-video-question-answer-on-activitynet-qaSlotNarrative–Accuracy: 50.3Confidence Score: 3.5
zeroshot-video-question-answer-on-msrvtt-qaSlotNarrative–Accuracy: 69.8Confidence Score: 3.4
zeroshot-video-question-answer-on-msvd-qaSlotNarrative–Accuracy: 75.6Confidence Score: 3.8