Side4Video: Spatial-Temporal Side Network for Memory-Efficient Image-to-Video Transfer Learning

Benchmark Model Rank Results
action-classification-on-kinetics-400Side4Video (EVA, ViT-E/14)#20Acc@1: 88.6Acc@5: 98.2
action-recognition-in-videos-on-somethingSide4Video (EVA ViT-E/14)#10Top-1 Accuracy: 75.2Top-5 Accuracy: 94.0
action-recognition-in-videos-on-something-1Side4Video (EVA ViT-E/14#3Top 1 Accuracy: 67.3Top 5 Accuracy: 88.8
video-retrieval-on-msr-vtt-1kaSide4Video#9text-to-video R@1: 52.3text-to-video R@5: 75.5
video-retrieval-on-msvdSide4Video#4text-to-video R@1: 56.1text-to-video R@5: 81.7
video-retrieval-on-vatexSide4Video#7text-to-video R@1: 68.8text-to-video R@5: 93.5