TS-LLaVA: Constructing Visual Tokens through Thumbnail-and-Sampling for Training-Free Video Large Language Models

Benchmark Model Rank Results
video-based-generative-performance-benchmarking-consistency-on-videoinstructTS-LLaVA-34B#2gpt-score: 3.69
zero-shot-video-question-answer-on-intentqaTS-LLaVA-34B#2Accuracy: 67.9
video-based-generative-performance-benchmarking-contextual-understanding-on-videoinstructTS-LLaVA-34B#3gpt-score: 3.86
video-based-generative-performance-benchmarking-correctness-of-information-on-videoinstructTS-LLaVA-34B#3gpt-score: 3.55
video-based-generative-performance-benchmarking-on-videoinstructTS-LLaVA-34B#3mean: 3.38
zeroshot-video-question-answer-on-tgif-qaTS-LLaVA-34B#3Accuracy: 81.0Confidence Score: 4.2
video-based-generative-performance-benchmarking-temporal-understanding-on-videoinstructTS-LLaVA-34B#5gpt-score: 2.77
video-based-generative-performance-benchmarking-detail-orientation-on-videoinstructTS-LLaVA-34B#6gpt-score: 3.03
zero-shot-video-question-answer-on-next-qaTS-LLaVA-34B#6Accuracy: 73.6
zeroshot-video-question-answer-on-activitynet-qaTS-LLaVA-34B#6Accuracy: 58.9Confidence Score: 3.5
zeroshot-video-question-answer-on-msrvtt-qaTS-LLaVA-34B#6Accuracy: 66.2Confidence Score: 3.6
zeroshot-video-question-answer-on-msvd-qaTS-LLaVA-34B#8Accuracy: 79.4Confidence Score: 4.1
zero-shot-video-question-answer-on-egoschema-subsetTS-LLaVA-34B#9Accuracy: 57.8