MoViNets: Mobile Video Networks for Efficient Video Recognition

Benchmark Model Rank Results
action-classification-on-charadesMoViNet-A6#3MAP: 63.2
action-classification-on-charadesMoViNet-A4#9MAP: 48.5
action-classification-on-charadesMoViNet-A2#26MAP: 32.5
action-classification-on-kinetics-400MoViNet-A6#73Acc@1: 81.5FLOPs (G) x views: 386x1
action-classification-on-kinetics-400MoViNet-A5#80Acc@1: 80.9Acc@5: 94.9FLOPs (G) x views: 281x1
action-classification-on-kinetics-400MoViNet-A4#86Acc@1: 80.5Acc@5: 94.5FLOPs (G) x views: 105x1
action-classification-on-kinetics-400MoViNet-A3#108Acc@1: 78.2Acc@5: 93.8FLOPs (G) x views: 56.9x1
action-classification-on-kinetics-400MoViNet-A2#136Acc@1: 75.0Acc@5: 92.3FLOPs (G) x views: 10.3x1
action-classification-on-kinetics-400MoViNet-A1#146Acc@1: 72.7Acc@5: 91.2FLOPs (G) x views: 6.0x1
action-classification-on-kinetics-400MoViNet-A0#163Acc@1: 65.8Acc@5: 87.4FLOPs (G) x views: 2.7x1
action-classification-on-kinetics-600MoViNet-A5 (AutoAugment)#27Top-1 Accuracy: 84.3Top-5 Accuracy: 96.4GFLOPs: 281x1
action-classification-on-kinetics-600MoViNet-A6#32Top-1 Accuracy: 83.5Top-5 Accuracy: 96.5GFLOPs: 386x1
action-classification-on-kinetics-600MoViNet-A5#36Top-1 Accuracy: 82.7Top-5 Accuracy: 95.7GFLOPs: 281x1
action-classification-on-kinetics-600MoViNet-A4#39Top-1 Accuracy: 81.2Top-5 Accuracy: 94.9GFLOPs: 105x1
action-classification-on-kinetics-600MoViNet-A3#41Top-1 Accuracy: 80.8Top-5 Accuracy: 80.8GFLOPs: 56.9x1
action-classification-on-kinetics-600MoViNet-A2#48Top-1 Accuracy: 77.5Top-5 Accuracy: 93.4GFLOPs: 10.3x1
action-classification-on-kinetics-600MoViNet-A1#50Top-1 Accuracy: 76.0Top-5 Accuracy: 92.6GFLOPs: 6.0x1
action-classification-on-kinetics-600MoViNet-A0#52Top-1 Accuracy: 71.5Top-5 Accuracy: 90.4GFLOPs: 2.7x1
action-classification-on-kinetics-700MoViNet-A6#18Top-1 Accuracy: 72.3
action-classification-on-kinetics-700MoViNet-A5#19Top-1 Accuracy: 71.7
action-classification-on-kinetics-700MoViNet-A4#20Top-1 Accuracy: 70.7
action-classification-on-kinetics-700MoViNet-A3#21Top-1 Accuracy: 68.0
action-classification-on-kinetics-700MoViNet-A2#22Top-1 Accuracy: 66.7
action-classification-on-kinetics-700MoViNet-A1#23Top-1 Accuracy: 63.5
action-classification-on-kinetics-700MoViNet-A0#24Top-1 Accuracy: 58.5
action-classification-on-moments-in-timeMoViNet-A6#6Top 1 Accuracy: 40.2
action-classification-on-moments-in-timeMoViNet-A5#7Top 1 Accuracy: 39.1
action-classification-on-moments-in-timeMoViNet-A4#8Top 1 Accuracy: 37.9
action-classification-on-moments-in-timeMoViNet-A3#11Top 1 Accuracy: 35.6
action-classification-on-moments-in-timeMoViNet-A2#12Top 1 Accuracy: 34.3
action-classification-on-moments-in-timeMoViNet-A1#16Top 1 Accuracy: 32.0
action-classification-on-moments-in-timeMoViNet-A0#23Top 1 Accuracy: 27.5
action-recognition-in-videos-on-somethingMoViNet-A2#85Top-1 Accuracy: 63.5Top-5 Accuracy: 89.0Parameters: 4.8M
action-recognition-in-videos-on-somethingMoViNet-A1#88Top-1 Accuracy: 62.7Top-5 Accuracy: 89.0Parameters: 4.6M
action-recognition-in-videos-on-somethingMoViNet-A0#94Top-1 Accuracy: 61.3Top-5 Accuracy: 88.2Parameters: 3.1M
action-recognition-in-videos-on-somethingMoViNet-A3#102Parameters: 5.3MGFLOPs: 23.7x1
action-recognition-on-epic-kitchens-100MoViNet-A6#11Action@1: 47.7Verb@1: 72.2Noun@1: 57.3GFLOPs: 117x1
action-recognition-on-epic-kitchens-100MoViNet-A5#14Action@1: 44.5Verb@1: 69.1Noun@1: 55.1GFLOPs: 74.9x1
action-recognition-on-epic-kitchens-100MoViNet-A4#17Action@1: 44.4Verb@1: 68.8Noun@1: 56.2GFLOPs: 42.2x1
action-recognition-on-epic-kitchens-100MoViNet-A2#22Action@1: 41.2Verb@1: 67.1Noun@1: 52.3GFLOPs: 7.59x1
action-recognition-on-epic-kitchens-100MoViNet-A0#25Action@1: 36.8Verb@1: 64.8Noun@1: 47.4GFLOPs: 1.74x1