VindLU: A Recipe for Effective Video-and-Language Pretraining

Benchmark Model Rank Results
video-question-answering-on-activitynet-qaVindLU#15Accuracy: 44.7
video-question-answering-on-msrvtt-mcVindLU#2Accuracy: 95.5
video-question-answering-on-msrvtt-qaVindLU#9Accuracy: 44.6
video-question-answering-on-tvqaVindLU#3Accuracy: 79.0
video-retrieval-on-activitynetVindLU#10text-to-video R@1: 55.0text-to-video R@5: 81.4
video-retrieval-on-condensed-moviesVINDLU#2text-to-video R@1: 18.4text-to-video R@5: 36.4
video-retrieval-on-didemoVindLU#8text-to-video R@1: 61.2text-to-video R@5: 85.8
video-retrieval-on-msr-vtt-1kaVindLU#28text-to-video R@1: 46.5text-to-video R@5: 71.5
video-retrieval-on-querydVINDLU#3text-to-video R@1: 67.8text-to-video R@5: 86.3
video-retrieval-on-ssv2-label-retrievalVindLU#3text-to-video R@1: 53.1text-to-video R@5: 81.8
video-retrieval-on-ssv2-template-retrievalVindLU#3text-to-video R@1: 83.3text-to-video R@10: 100text-to-video R@5: 100