Video Swin Transformer

Benchmark Model Rank Results
action-classification-on-kinetics-400Swin-L (384x384, ImageNet-21k pretrain)#54Acc@1: 84.9Acc@5: 96.7
action-classification-on-kinetics-400Swin-L (ImageNet-21k pretrain)#61Acc@1: 83.1Acc@5: 95.9
action-classification-on-kinetics-400Swin-B (ImageNet-21k pretrain)#66Acc@1: 82.7Acc@5: 95.5
action-classification-on-kinetics-400Swin-B (ImageNet-1k pretrain)#83Acc@1: 80.6Acc@5: 94.6
action-classification-on-kinetics-400Swin-S (ImageNet-1k pretrain)#84Acc@1: 80.6Acc@5: 94.5
action-classification-on-kinetics-400Swin-T (ImageNet-1k pretrain)#105Acc@1: 78.8Acc@5: 93.6
action-classification-on-kinetics-600Swin-L (384x384, ImageNet-21k pretrain)#22Top-1 Accuracy: 86.1Top-5 Accuracy: 97.3
action-classification-on-kinetics-600Swin-B (ImageNet-21k pretrain)#29Top-1 Accuracy: 84.0Top-5 Accuracy: 96.5
action-recognition-in-videos-on-somethingSwin-B (IN-21K + Kinetics400 pretrain)#39Top-1 Accuracy: 69.6Top-5 Accuracy: 92.7Parameters: 89