TAda! Temporally-Adaptive Convolutions for Video Understanding

Open paper
Benchmark Model Rank Results
action-classification-on-kinetics-400TAdaConvNeXt-T#102Acc@1: 79.1Acc@5: 93.7
action-classification-on-kinetics-400TAda2D-En (ResNet-50, 8+16 frames)#109Acc@1: 78.2Acc@5: 93.5
action-classification-on-kinetics-400TAda2D (ResNet-50, 16 frames)#122Acc@1: 77.4Acc@5: 93.1
action-classification-on-kinetics-400TAda2D (ResNet-50, 8 frames)#126Acc@1: 76.7Acc@5: 92.6
action-recognition-in-videos-on-somethingTAda2D-En (ResNet-50, 8+16 frames)#60Top-1 Accuracy: 67.2Top-5 Accuracy: 89.8
action-recognition-in-videos-on-somethingTAdaConvNeXt-T#62Top-1 Accuracy: 67.1Top-5 Accuracy: 90.4
action-recognition-in-videos-on-somethingTAda2D (ResNet-50, 16 frames)#75Top-1 Accuracy: 65.6Top-5 Accuracy: 89.2
action-recognition-in-videos-on-somethingTAda2D (ResNet-50, 8 frames)#84Top-1 Accuracy: 64.0Top-5 Accuracy: 88.0

Rank counts only results with a code link.