Surface Normal Prediction on Taskonomy (test)
87.3Surface Normal AccuracyAutoMTL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AutoMTL# Params (M)=53.1062021.10 | 87.3 | — | — | — | — | |
| Learn to Branch# Params (M)=30.6502021.10 | 85 | — | — | — | — | |
| Single-Task# Params (M)=106.4242021.10 | 80.7 | — | — | — | — | |
| AdaShare# Params (M)=21.2852021.10 | 80.2 | — | — | — | — | |
| NDDR-CNN# Params (M)=115.1512021.10 | 80 | — | — | — | — | |
| Multi-Task# Params (M)=21.2852021.10 | 79.6 | — | — | — | — | |
| Sluice# Params (M)=106.4242021.10 | 79.5 | — | — | — | — | |
| MTAN# Params (M)=95.9942021.10 | 78.7 | — | — | — | — | |
| DEN# Params (M)=23.8382021.10 | 78.6 | — | — | — | — | |
| Cross-Stitch# Params (M)=106.4242021.10 | 77.9 | — | — | — | — | |
| DeepLabv3+backbone=ResNet-502023.11 | — | 11.45 | 4.938 | 19.65 | 56.13 | |
| PolyMaXbackbone=ResNet-502023.11 | — | 10.96 | 4.922 | 18.83 | 55.27 | |
| PolyMaXbackbone=ConvNeXt-T2023.11 | — | 10.87 | 4.859 | 18.49 | 55.39 | |
| PolyMaXbackbone=ConvNeXt-S2023.11 | — | 10.67 | 4.709 | 18.46 | 56.05 | |
| PolyMaXbackbone=ConvNeXt-B2023.11 | — | 10.58 | 4.646 | 18.37 | 56.43 | |
| PolyMaXbackbone=ConvNeXt-L2023.11 | — | 10.46 | 4.55 | 18.25 | 56.76 |