Human Attribute Prediction on WIDER-Attribute (test)
90.1mAPViT-L16 + CSRA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ViT-L16 + CSRABackbone=ViT-L16, Module=CSRA2021.08 | 90.1 | 81 | 85.2 | |
| ViT-B16 + CSRABackbone=ViT-B16, Module=CSRA2021.08 | 89 | 79.4 | 84.3 | |
| ViT-L16Backbone=ViT-L162021.08 | 87.7 | 78.1 | 82.8 | |
| VAC2021.08 | 87.5 | 77.6 | 82.4 | |
| VAA2021.08 | 86.4 | — | — | |
| ViT-B16Backbone=ViT-B162021.08 | 86.3 | 75.9 | 81.5 | |
| ResNet-SRNBackbone=ResNet-101, Variant=Full SRN2017.02 | 86.2 | 75.9 | 81.3 | |
| SRN2021.08 | 86.2 | 75.9 | 81.3 | |
| ResNet-SRN-attBackbone=ResNet-101, Variant=Attention-based2017.02 | 85.4 | 74.9 | 80.8 | |
| ResNet-101-semanticBackbone=ResNet-101, Mechanism=Label semantic relations2017.02 | 85.1 | 74.8 | 80.5 | |
| ResNet-101Backbone=ResNet-1012017.02 | 85 | 74.7 | 80.4 | |
| ResNet-107Backbone=ResNet-1072017.02 | 85 | 74.8 | 80.6 | |
| VA2021.08 | 82.9 | — | — | |
| DHCInput=Full image and bounding box, Supervision=Includes event labels2017.02 | 81.3 | — | — | |
| DHC2021.08 | 81.3 | — | — | |
| R*CNNInput=Full image and bounding box2017.02 | 80.5 | — | — | |
| R-CNNInput=Cropped human patches2017.02 | 80 | — | — |