8 Grouped Facial Attribute Classification on CelebA
0.727PrecisionETR-NLP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ETR-NLPBackbone=ResNet18, #P (M)=82023.08 | 0.727 | 0.648 | 0.678 | 7.8 | |
| Task routingBackbone=ResNet18, #P (M)=11.22023.08 | 0.717 | 0.617 | 0.655 | 4.4 | |
| Max roamingBackbone=ResNet18, #P (M)=11.22023.08 | 0.712 | 0.63 | 0.662 | 5.2 | |
| Atten. hard sharingBackbone=ResNet18, #P (M)=12.92023.08 | 0.711 | 0.626 | 0.659 | 4.8 | |
| GradNormBackbone=ResNet18, #P (M)=11.2, alpha=0.52023.08 | 0.704 | 0.595 | 0.636 | 1.5 | |
| MGDA-UBBackbone=ResNet18, #P (M)=11.22023.08 | 0.686 | 0.602 | 0.636 | 1 | |
| Hard sharingBackbone=ResNet18, #P (M)=11.22023.08 | 0.677 | 0.598 | 0.63 | 0 |