Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers

Benchmark Model Rank Results
action-classification-on-kinetics-400Motionformer-HR#77Acc@1: 81.1Acc@5: 95.2
action-recognition-in-videos-on-somethingMformer-L#47Top-1 Accuracy: 68.1Top-5 Accuracy: 91.2Parameters: N/A
action-recognition-in-videos-on-somethingMformer-HR#61Top-1 Accuracy: 67.1Top-5 Accuracy: 90.6Parameters: N/A
action-recognition-in-videos-on-somethingMformer#68Top-1 Accuracy: 66.5Top-5 Accuracy: 90.1
action-recognition-on-epic-kitchens-100Mformer-HR#15Action@1: 44.5Verb@1: 67.0Noun@1: 58.5
action-recognition-on-epic-kitchens-100Mformer-L#18Action@1: 44.1Verb@1: 67.1Noun@1: 57.6
action-recognition-on-epic-kitchens-100Mformer#21Action@1: 43.1Verb@1: 66.7Noun@1: 56.5