MotionSqueeze: Neural Motion Feature Learning for Video Understanding

Open paper
Benchmark Model Rank Results
action-classification-on-kinetics-400MSNet-R50 (16 frames, ImageNet pretrained)#128Acc@1: 76.4
action-recognition-in-videos-on-hmdb-51MSNet-R50 (16 frames, ImageNet pretrained)#19Average accuracy of 3 splits: 77.4
action-recognition-in-videos-on-somethingMSNet-R50En (8+16 ensemble, ImageNet pretrained)#66Top-1 Accuracy: 66.6Top-5 Accuracy: 90.6
action-recognition-in-videos-on-somethingMSNet-R50 (16 frames, ImageNet pretrained)#81Top-1 Accuracy: 64.7Top-5 Accuracy: 89.4
action-recognition-in-videos-on-somethingMSNet-R50 (8 frames, ImageNet pretrained)#87Top-1 Accuracy: 63Top-5 Accuracy: 88.4
action-recognition-in-videos-on-something-1MSNet-R50En (ensemble)#24Top 1 Accuracy: 55.1
action-recognition-in-videos-on-something-1MSNet-R50En (8+16 ensemble, ImageNet pretrained)#26Top 1 Accuracy: 54.4Top 5 Accuracy: 83.8
action-recognition-in-videos-on-something-1MSNet-R50 (16 frames, ImageNet pretrained)#36Top 1 Accuracy: 52.1Top 5 Accuracy: 82.3
action-recognition-in-videos-on-something-1MSNet-R50 (8 frames, ImageNet pretrained)#40Top 1 Accuracy: 50.9Top 5 Accuracy: 80.3

Rank counts only results with a code link.