Action Classification on Kinetics 400 2017 (test)
80.6Top-1 AccuracyViViT-L
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ViViT-LPretrain=IN-21K, Frames=16×3×4, GFLOPs=14462021.10 | 80.6 | 94.6 | |
| MViT-BPretrain=-, Frames=32×1×5, GFLOPs=1702021.10 | 80.2 | 94.4 | |
| TAdaConvNeXt-TPretrain=IN-1K, Frames=32×3×4, GFLOPs=942021.10 | 79.1 | 93.7 | |
| Swin-TPretrain=IN-1K, Frames=32×3×4, GFLOPs=882021.10 | 78.8 | 93.6 | |
| MViT-BPretrain=-, Frames=16×1×5, GFLOPs=70.52021.10 | 78.4 | 93.5 | |
| TAdaConvNeXt-TPretrain=IN-1K, Frames=16×3×4, GFLOPs=472021.10 | 78.4 | 93.5 | |
| TAda2D_EnPretrain=IN-1K, Frames=(8+16)×3×10, GFLOPs=1292021.10 | 78.2 | 93.5 | |
| TimeSformerPretrain=IN-21K, Frames=8×3×1, GFLOPs=1962021.10 | 78 | 93.7 | |
| TDNPretrain=IN-1K, Frames=(16+64)×3×10, GFLOPs=942021.10 | 77.5 | 93.2 | |
| TAda2DPretrain=IN-1K, Frames=16×3×10, GFLOPs=862021.10 | 77.4 | 93.1 | |
| CorrNetPretrain=IN-1K, Frames=32×1×10, GFLOPs=1152021.10 | 77.2 | — | |
| SlowFast 8×8Pretrain=-, Frames=(8+32)×3×10, GFLOPs=65.72021.10 | 77 | 92.6 | |
| TANetPretrain=IN-1K, Frames=16×3×10, GFLOPs=862021.10 | 76.9 | 92.9 | |
| TAda2DPretrain=IN-1K, Frames=8×3×10, GFLOPs=432021.10 | 76.7 | 92.6 | |
| TDNPretrain=IN-1K, Frames=(8+32)×3×10, GFLOPs=472021.10 | 76.6 | 92.8 | |
| SmallBigNetPretrain=IN-1K, Frames=8×3×10, GFLOPs=572021.10 | 76.3 | 92.5 | |
| TANetPretrain=IN-1K, Frames=8×3×10, GFLOPs=432021.10 | 76.3 | 92.6 | |
| SlowFast 4×16Pretrain=-, Frames=(4+32)×3×10, GFLOPs=36.12021.10 | 75.6 | 92.1 | |
| TSMPretrain=IN-1K, Frames=8×3×10, GFLOPs=432021.10 | 74.1 | — |