Video Recognition on Kinetics-600 11 (val)
83.8Top-1 AccMViT-B-24, 32×3
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MViT-B-24, 32×3pre-train=from-scratch, GFLOPs x views=236 × 1 × 5, Param=52.9M2021.04 | 83.8 | 96.3 | |
| MViT-B, 32×3pre-train=from-scratch, GFLOPs x views=170 × 1 × 5, Param=36.8M2021.04 | 83.4 | 96.3 | |
| ViT-L-ViViTpre-train=IN-21K, GFLOPs x views=3992 × 3 × 4, Param=310.8M2021.04 | 83 | 95.7 | |
| ViT-B-TimeSformerpre-train=IN-21K, GFLOPs x views=1703 × 3 × 1, Param=121.4M2021.04 | 82.4 | 96 | |
| MViT-B, 16×4pre-train=from-scratch, GFLOPs x views=70.5 × 1 × 5, Param=36.8M2021.04 | 82.1 | 95.7 | |
| X3D-XLGFLOPs x views=48.4 × 3 × 10, Param=11.0M2021.04 | 81.9 | 95.5 | |
| SlowFast 16×8+NLGFLOPs x views=234 × 3 × 10, Param=59.9M2021.04 | 81.8 | 95.1 | |
| X3D-MGFLOPs x views=6.2 × 3 × 10, Param=3.8M2021.04 | 78.8 | 94.5 |