Action Recognition on AVA v2.1 (val)
27.8mAPAVSlowFast
Evaluation results
| Method | Links | |
|---|---|---|
| AVSlowFastinput=V+A, pre-train=K4002021.06 | 27.8 | |
| LFBinput=V, pre-train=K4002021.06 | 27.7 | |
| Action Transformer (Tx+I3D+96f)Modalities=RGB, Architecture=I3D, Transformer, Input Frames=96f2018.12 | 25 | |
| Girdhar et al.input=V, pre-train=K4002021.06 | 25 | |
| Action Transformer (Tx+I3D head)Modalities=RGB, Architecture=I3D, Transformer2018.12 | 24.9 | |
| Action Transformer (Tx-only head)Modalities=RGB, Architecture=I3D, Transformer2018.12 | 24.4 | |
| RTPRinput=V+F, pre-train=ImageNet2021.06 | 22.3 | |
| Zhang et al.input=V, pre-train=K4002021.06 | 22.2 | |
| STEPinput=V, pre-train=K4002021.06 | 18.6 | |
| ARCNModalities=RGB, Flow, Architecture=S3D-G, Relation Nets2018.12 | 17.4 | |
| ACRNinput=V+F, pre-train=K4002021.06 | 17.4 | |
| AVA baselineModalities=RGB, Flow, Architecture=I3D, Faster R-CNN, ResNet-502018.12 | 15.6 | |
| AVAinput=V+F, pre-train=K4002021.06 | 15.6 | |
| Single frameModalities=RGB, Flow, Architecture=ResNet-50, Faster R-CNN2018.12 | 14.7 |