Video Classification on Kinetics-600 v1.0 (test)
88.3Top-1 AccuracyMaskFeat (MViT-L↑312, 40×3)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MaskFeat (MViT-L↑312, 40×3)pre-train=MaskFeat, K600, FLOPs×views=2828×3×4, Param=218M2021.12 | 88.3 | 98 | |
| Florence 384pre-train=Text, FLD-900M, FLOPs×views=N/A×3×4, Param=647M2021.12 | 87.8 | 97.8 | |
| MViT-L↑312, 40×3pre-train=Sup., IN-21K, FLOPs×views=2828×3×4, Param=218M2021.12 | 87.5 | 97.8 | |
| MaskFeat (MViT-L, 16×4)pre-train=MaskFeat, K600, FLOPs×views=377×1×10, Param=218M2021.12 | 86.4 | 97.4 | |
| Swin-L↑384, 32×2pre-train=Sup., IN-21K, FLOPs×views=2107×5×10, Param=200M2021.12 | 86.1 | 97.3 | |
| ViViT-Hpre-train=Sup., JFT-300M, FLOPs×views=3981×3×4, Param=654M2021.12 | 85.8 | 96.5 | |
| MViT-L, 16×4pre-train=Sup., IN-21K, FLOPs×views=377×1×10, Param=218M2021.12 | 85.8 | 97.1 | |
| MoViNet-A6pre-train=-, FLOPs×views=386×1×1, Param=31M2021.12 | 84.8 | 96.5 | |
| MViTv1-B-24, 32×3pre-train=-, FLOPs×views=236×1×5, Param=53M2021.12 | 84.1 | 96.5 | |
| Swin-B, 16×2pre-train=Sup., IN-21K, FLOPs×views=282×3×4, Param=88M2021.12 | 84 | 96.5 | |
| X3D-XLpre-train=-, FLOPs×views=48×3×10, Param=11M2021.12 | 81.9 | 95.5 | |
| SlowFast 16×8+NLpre-train=-, FLOPs×views=234×3×10, Param=60M2021.12 | 81.8 | 95.1 |