Is Space-Time Attention All You Need for Video Understanding?

Benchmark Model Rank Results
action-classification-on-kinetics-400TimeSformer-L#82Acc@1: 80.7Acc@5: 94.7FLOPs (G) x views: 7140x3
action-classification-on-kinetics-400TimeSformer-HR#95Acc@1: 79.7Acc@5: 94.4
action-classification-on-kinetics-400TimeSformer#111Acc@1: 78Acc@5: 93.7
action-recognition-in-videos-on-somethingTimeSformer-HR#89Top-1 Accuracy: 62.5
action-recognition-in-videos-on-somethingTimeSformer-L#90Top-1 Accuracy: 62.3
action-recognition-in-videos-on-somethingTimeSformer#95Top-1 Accuracy: 59.5
action-recognition-on-diving-48TimeSformer-L#11Accuracy: 81
action-recognition-on-diving-48TimeSformer-HR#12Accuracy: 78
action-recognition-on-diving-48TimeSformer#14Accuracy: 75
anomaly-detection-on-ubnormalTimeSformer#5AUC: 68.5%RBDC: 0.04TBDC: 0.05