Temporally-Adaptive Models for Efficient Video Understanding

Benchmark Model Rank Results
action-classification-on-kinetics-400TAdaFormer-L/14#12Acc@1: 89.9
action-classification-on-kinetics-400TAdaConvNeXtV2-B#42Acc@1: 86.4
action-recognition-in-videos-on-somethingTAdaFormer-L/14#19Top-1 Accuracy: 73.6
action-recognition-in-videos-on-somethingTAdaConvNeXtV2-B#30Top-1 Accuracy: 71.1
action-recognition-in-videos-on-something-1TAdaFormer-L/14#5Top 1 Accuracy: 63.7
action-recognition-in-videos-on-something-1TAdaConvNeXtV2-B#9Top 1 Accuracy: 60.7
action-recognition-on-epic-kitchens-100TAdaFormer-L/14#4Action@1: 51.8Verb@1: 71.7Noun@1: 64.1
action-recognition-on-epic-kitchens-100TAdaConvNeXtV2-S#9Action@1: 48.9Verb@1: 71.0Noun@1: 60.2