Swin-L (384x384, ImageNet-21k pretrain)

Reported on 4 benchmarks across 1 task · 1 paper · 2 SOTA

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Computer Vision4 results

VideoonKinetics-400
Acc@5· 2021-06-24
96.7
best: 98.9 (TubeViT-H (ImageNet-1k))
SOTA
Video Swin Transformer arXiv:2106.13230
VideoonKinetics-600
Top-5 Accuracy· uses extra data· 2021-06-24
97.3
best: 98.9 (TubeVit-H)
SOTA
Video Swin Transformer arXiv:2106.13230
VideoonKinetics-400
Acc@1· 2021-06-24
84.9
best: 93.6 (OmniVec2)
Video Swin Transformer arXiv:2106.13230
VideoonKinetics-600
Top-1 Accuracy· uses extra data· 2021-06-24
86.1
best: 91.9 (InternVideo2-6B)
Video Swin Transformer arXiv:2106.13230