Iwin Transformer: Hierarchical Vision Transformer using Interleaved Windows

Benchmark Model Rank Results
action-classification-on-kinetics-400Iwin-S k133 w1677#98Acc@1: 80.0Acc@5: 94.1
action-classification-on-kinetics-400Iwin-T k133 w1677#111Acc@1: 79.1Acc@5: 93.8
action-classification-on-kinetics-400Iwin-T k233 w477#115Acc@1: 78.8Acc@5: 93.8
action-classification-on-kinetics-400Iwin-T k233 w877#116Acc@1: 78.7Acc@5: 93.8
action-classification-on-kinetics-400Iwin-T k233 w277#122Acc@1: 78.2Acc@5: 93.8
image-generation-on-imagenet-256x256FlashDiT(win8)#123FID: 2.30
image-generation-on-imagenet-256x256FlashDiT(win4)#149FID: 2.90