TDN: Temporal Difference Networks for Efficient Action Recognition

Benchmark Model Rank Results
action-recognition-in-videos-on-something-something-v1TDN ResNet101 (one clip, center crop, 8+16 ensemble, ImageNet pretrained, RGB only)#16Top 1 Accuracy: 56.8Top 5 Accuracy: 84.1
action-recognition-in-videos-on-something-something-v2TDN ResNet101 (one clip, three crop, 8+16 ensemble, ImageNet pretrained, RGB only)#40Top-1 Accuracy: 69.6Top-5 Accuracy: 92.2GFLOPs: 198x3
action-recognition-in-videos-on-something-something-v2TDN ResNet101 (one clip, center crop, 8+16 ensemble, ImageNet pretrained, RGB only)#48Top-1 Accuracy: 68.2Top-5 Accuracy: 91.6GFLOPs: 198x1
action-classification-on-kinetics-400TDN-ResNet101 (ensemble, ImageNet pretrained, RGB only)#104Acc@1: 79.4Acc@5: 94.4