Video Classification on Kinetics-700 (val)
81.1Top-1 AccHiera-H
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Hiera-Hbackbone=Hiera-H, pretrain=MAE, FLOPs (G)=1159×3×5, Param=672M2023.06 | 81.1 | — | |
| MViT-L↑312, 40x3pre-train=MaskFeat, K700, FLOPs x views=2828x3x4, Param=2182021.12 | 80.4 | 95.7 | |
| Hiera-Lbackbone=Hiera-L, pretrain=MAE, FLOPs (G)=413×3×5, Param=213M2023.06 | 80.3 | — | |
| MVIT-L↑312, 40x3pre-train=Sup., IN-21K, FLOPs x views=2828x3x4, Param=2182021.12 | 79.4 | 94.9 | |
| MVIT-L, 16x4pre-train=MaskFeat, K700, FLOPs x views=377x1x10, Param=2182021.12 | 77.5 | 93.8 | |
| MViTv2-Lbackbone=MViTv2-L, pretrain=MaskFeat, FLOPs (G)=377×1×10, Param=218M2023.06 | 77.5 | — | |
| MVIT-L, 16x4pre-train=Sup., IN-21K, FLOPs x views=377x1x10, Param=2182021.12 | 76.7 | 93.4 | |
| MViTv2-Lbackbone=MViTv2-L, pretrain=Sup, IN-21K, FLOPs (G)=377×1×10, Param=218M2023.06 | 76.7 | — | |
| MoViNet-A6pre-train=N/A, FLOPs x views=386x1x1, Param=312021.12 | 72.3 | — | |
| SlowFast 16x8+NLpre-train=N/A, FLOPs x views=234x3x10, Param=602021.12 | 71 | 89.6 | |
| EVA CLIP-gzero-shot evaluation protocol=True2022.11 | 58.4 | — | |
| OpenAI CLIP-Lzero-shot evaluation protocol=True2022.11 | 57.7 | — | |
| Open CLIP-Hzero-shot evaluation protocol=True2022.11 | 56.1 | — | |
| Open CLIP-gzero-shot evaluation protocol=True2022.11 | 55 | — |