Multi-task Learning on NYU LibMTL v2 (test)
54.59Segmentation ScoreHPS + GLS
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| HPS + GLSBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 54.59 | 0.3785 | 22.71 | 1.97 | |
| HPS + UWBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 54.29 | 0.3815 | 23.48 | 0.44 | |
| HPS + GradVacBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 54.21 | 0.3859 | 23.58 | -0.14 | |
| HPS + DWABackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 54.06 | 0.382 | 23.7 | -0.06 | |
| HPS + RLWBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 54.04 | 0.3827 | 23.07 | 0.76 | |
| HPS + CAGradBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.97 | 0.3885 | 22.47 | 1.06 | |
| HPS + PCGradBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.94 | 0.3804 | 23.52 | 0.26 | |
| HPSBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.93 | 0.3825 | 23.57 | 0 | |
| HPS + GradNormBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.91 | 0.3842 | 23.17 | 0.41 | |
| TSσBNBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.78 | 0.3735 | 22.3 | 2.48 | |
| HPS + GradDropBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.73 | 0.3837 | 23.54 | -0.19 | |
| HPS + IMTLBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.63 | 0.3868 | 22.58 | 0.84 | |
| HPS + MGDABackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.52 | 0.3852 | 22.74 | 0.69 | |
| HPS + Nash-MTLBackbone=DeepLabV3–ResNet50, Architecture=HPS, Pre-trained=true2025.12 | 53.41 | 0.3867 | 22.57 | 0.73 |