Temporal Action Detection on THUMOS14 (test)
69.6mAPVideoMAE V2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VideoMAE V2Optical Flow=false2023.03 | 69.6 | — | |
| ViT-TADBackbone=ViT-B, Frame rate=8 FPS, Frame resolution=160x160, GPU=NVIDIA TITAN Xp2023.12 | 69.5 | 845 | |
| TriDetBackbone=ViT-B, Frame rate=30 FPS, Frame resolution=224x224, GPU=NVIDIA GeForce RTX 30902023.12 | 68.2 | 239 | |
| ActionFormerOptical Flow=true2023.03 | 66.8 | — | |
| ActionFormerBackbone=ViT-B, Frame rate=30 FPS, Frame resolution=224x224, GPU=NVIDIA GeForce RTX 30902023.12 | 65.1 | 268 | |
| ViT-TADBackbone=ViT-S, Frame resolution=160x160, GPU=NVIDIA TITAN Xp2023.12 | 64.3 | 1,349 | |
| ViT-TADBackbone=ViT-S, Frame resolution=112x112, GPU=NVIDIA TITAN Xp2023.12 | 63 | 2,135 | |
| BasicTADOptical Flow=false2023.03 | 59.6 | — | |
| TALLFormerOptical Flow=false2023.03 | 59.2 | — | |
| Explicit Dropout (V)Dropout on Attention Sequence=Explicit (V), Dropout on FF network=Explicit2026.04 | 56.51 | — | |
| Explicit Dropout (AV)Dropout on Attention Sequence=Explicit (AV), Dropout on FF network=Explicit2026.04 | 56.1 | — | |
| DropKeyDropout on Attention Sequence=DropKey [20], Dropout on FF network=Implicit2026.04 | 55.87 | — | |
| Implicit BaselineDropout on Attention Sequence=None, Dropout on FF network=Implicit2026.04 | 55.62 | — | |
| Explicit Dropout (Q)Dropout on Attention Sequence=Explicit (Q), Dropout on FF network=Explicit2026.04 | 54.98 | — | |
| Explicit Dropout (FF only)Dropout on Attention Sequence=None, Dropout on FF network=Explicit2026.04 | 54.92 | — | |
| Explicit Dropout (K)Dropout on Attention Sequence=Explicit (K), Dropout on FF network=Explicit2026.04 | 54.92 | — | |
| PlusTADModality=RGB, Temporal Resolution=96, Sampling Frame Rate=3 FPS, Spatial Resolution=short-128, Backbone=TP-SlowOnly(x1), Neck=SP-NECK, Head=Anchor-based2022.05 | 54.9 | 17,454 | |
| PlusTADModality=RGB, Temporal Resolution=96, Sampling Frame Rate=3 FPS, Spatial Resolution=short-128, Backbone=TP-SlowOnly(x1), Neck=SP-NECK, Head=Anchor-free2022.05 | 54.5 | 8,377 | |
| BasicTADBackbone=SlowOnly, Frame resolution=112x112, GPU=NVIDIA TITAN Xp2023.12 | 54.5 | 7,143 | |
| TadTROptical Flow=true2023.03 | 54.2 | — | |
| e2e-tadtrBackbone=ResNet-50-SlowFast, GPU=NVIDIA TITAN Xp2023.12 | 54.2 | 5,076 | |
| DropAttentionDropout on Attention Sequence=DropAttention [23], Dropout on FF network=Implicit2026.04 | 53.89 | — | |
| DCANOptical Flow=true2023.03 | 52.3 | — | |
| AFSDOptical Flow=true2023.03 | 52 | — | |
| AFSDBackbone=I3D, GPU=NVIDIA GeForce GTX 1080Ti2023.12 | 52 | 3,259 | |
| BasicTADModality=RGB, Temporal Resolution=768, Sampling Frame Rate=24 FPS, Spatial Resolution=128x128, Backbone=SlowOnly(x8), Neck=TDM, Head=Anchor-free2022.05 | 50.5 | 2,702 | |
| BasicTADModality=RGB, Temporal Resolution=768, Sampling Frame Rate=24 FPS, Spatial Resolution=128x128, Backbone=SlowOnly(x8), Neck=TDM+TFPN, Head=Anchor-based2022.05 | 50.2 | 5,454 | |
| DaoTADModality=RGB, Temporal Resolution=768, Sampling Frame Rate=25 FPS, Spatial Resolution=128x128, Backbone=R50-I3D(x8), Neck=TDM+TFPN, Head=RetinaNet2022.05 | 50 | 6,668 | |
| DaoTADOptical Flow=false2023.03 | 50 | — | |
| Arora et al. (FF)Dropout on Attention Sequence=None, Dropout on FF network=Arora et al. [1]2026.04 | 48.44 | — | |
| PBRNetBackbone=I3D, GPU=NVIDIA GeForce GTX 1080Ti2023.12 | 47.1 | 1,488 | |
| Arora et al. (Q)Dropout on Attention Sequence=(Q) Arora et al. [1], Dropout on FF network=Explicit2026.04 | 46.67 | — | |
| Arora et al. (K)Dropout on Attention Sequence=(K) Arora et al. [1], Dropout on FF network=Explicit2026.04 | 44.52 | — | |
| AFSDModality=RGB, Temporal Resolution=256, Sampling Frame Rate=10 FPS, Spatial Resolution=112×112, Backbone=I3D(x8), Neck=TFPN, Head=AFSD2022.05 | 43.8 | 4,057 | |
| RTD-NetOptical Flow=true2023.03 | 43.6 | — | |
| Arora et al. (V)Dropout on Attention Sequence=(V) Arora et al. [1], Dropout on FF network=Explicit2026.04 | 36.96 | — | |
| R-C3DModality=RGB, Temporal Resolution=768, Sampling Frame Rate=25 FPS, Spatial Resolution=112×112, Backbone=C3D(x8), Neck=TDM, Head=R-C3D2022.05 | 36.4 | 1,030 |