Multi-Task Learning on NYU v2
54.64mIoUMTAN
Evaluation results
| Method | Links | |||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MTANBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.28, #F (Relative FLOPs)=1.562025.12 | 54.64 | — | — | — | — | — | 23.12 | — | — | — | — | — | 0.3771 | — | — | — | — | — | — | 1.55 | — | |
| MoCo2023.05 | 54.05 | — | — | — | — | — | 23.39 | 16.69 | 35.65 | 61.68 | 72.6 | 0.3812 | 0.153 | — | — | — | 75.58 | -1.47 | 0.18 | — | — | |
| HPSBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.00, #F (Relative FLOPs)=1.002025.12 | 53.93 | — | — | — | — | — | 23.57 | — | — | — | — | — | 0.3825 | — | — | — | — | — | — | 0 | — | |
| RGW2023.05 | 53.85 | — | — | — | — | — | 23.67 | 17.24 | 34.62 | 60.49 | 71.75 | 0.3772 | 0.1562 | — | — | — | 75.87 | -0.62 | 1.03 | — | — | |
| TSσBNBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.00, #F (Relative FLOPs)=1.692025.12 | 53.78 | — | — | — | — | — | 22.31 | — | — | — | — | — | 0.3735 | — | — | — | — | — | — | 2.48 | — | |
| Static (EW)2023.05 | 53.77 | — | — | — | — | — | 23.57 | 17.04 | 35.04 | 60.93 | 72.07 | 0.3845 | 0.1605 | — | — | — | 75.45 | 0 | 1.63 | — | — | |
| DSelect-kBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.38, #F (Relative FLOPs)=1.342025.12 | 53.75 | — | — | — | — | — | 23.18 | — | — | — | — | — | 0.3802 | — | — | — | — | — | — | 0.64 | — | |
| PCGrad2023.05 | 53.7 | — | — | — | — | — | 23.43 | 16.97 | 35.16 | 61.19 | 72.28 | 0.3903 | 0.1607 | — | — | — | 75.41 | 0.16 | 1.79 | — | — | |
| GradNorm2023.05 | 53.58 | — | — | — | — | — | 23.44 | 16.98 | 35.11 | 61.11 | 72.24 | 0.3931 | 0.1663 | — | — | — | 75.06 | 0.99 | 2.62 | — | — | |
| CSBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.65, #F (Relative FLOPs)=1.692025.12 | 53.44 | — | — | — | — | — | 23.15 | — | — | — | — | — | 0.3818 | — | — | — | — | — | — | 0.35 | — | |
| TSBNBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.00, #F (Relative FLOPs)=1.692025.12 | 53.44 | — | — | — | — | — | 23.01 | — | — | — | — | — | 0.3761 | — | — | — | — | — | — | 1.04 | — | |
| MoDooptimizer=Adam, xt step size (at)=2.5 x 10^-4, vt step size (vt)=10^-3, weight decay=10^-5, batch size=82023.05 | 53.37 | — | — | — | — | — | 23.22 | 16.65 | 35.62 | 61.84 | 72.76 | 0.3739 | 0.1531 | — | — | — | 75.25 | -1.59 | 0.07 | — | — | |
| CGCBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=2.01, #F (Relative FLOPs)=2.032025.12 | 53.27 | — | — | — | — | — | 22.14 | — | — | — | — | — | 0.3914 | — | — | — | — | — | — | 0.84 | — | |
| MMOEBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.35, #F (Relative FLOPs)=1.342025.12 | 53.14 | — | — | — | — | — | 23.02 | — | — | — | — | — | 0.3876 | — | — | — | — | — | — | -0.15 | — | |
| CAGrad2023.05 | 53.12 | — | — | — | — | — | 22.53 | 15.88 | 37.42 | 63.5 | 74.17 | 0.3871 | 0.1599 | — | — | — | 75.19 | -1.36 | 0.26 | — | — | |
| PLEBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=2.41, #F (Relative FLOPs)=2.712025.12 | 52.75 | — | — | — | — | — | 22.1 | — | — | — | — | — | 0.3943 | — | — | — | — | — | — | 0.32 | — | |
| LTBBackbone=ResNet-50, Architecture Framework=LibMTL / DeepLabV3, #P (Relative Parameters)=1.65, #F (Relative FLOPs)=1.692025.12 | 52.58 | — | — | — | — | — | 23.31 | — | — | — | — | — | 0.3828 | — | — | — | — | — | — | -0.49 | — | |
| MGDA-UB2023.05 | 50.42 | — | — | — | — | — | 22.78 | 16.14 | 36.9 | 62.88 | 73.61 | 0.3834 | 0.1555 | — | — | — | 73.46 | -0.38 | 1.26 | — | — | |
| MARIGOLD2026.03 | 41.01 | — | — | — | — | — | 25.29 | 20.06 | 28.55 | 55.51 | 68.11 | 0.527 | 0.2137 | — | — | — | 66.27 | — | — | -4.54 | 1 | |
| LS2026.03 | 39.29 | — | — | — | — | — | 28.15 | 23.96 | 22.09 | 47.5 | 61.08 | 0.5493 | 0.2263 | — | — | — | 65.33 | — | — | 5.59 | 4.22 | |
| DWA2026.03 | 39.11 | — | — | — | — | — | 27.61 | 23.18 | 24.17 | 50.18 | 62.39 | 0.551 | 0.2285 | — | — | — | 65.31 | — | — | 3.57 | 3.33 | |
| SI2026.03 | 38.45 | — | — | — | — | — | 27.6 | 23.37 | 22.53 | 48.57 | 62.32 | 0.5354 | 0.2201 | — | — | — | 64.27 | — | — | 4.39 | 3.44 | |
| STL2026.03 | 38.3 | — | — | — | — | — | 25.01 | 19.21 | 30.14 | 57.2 | 69.15 | 0.6754 | 0.278 | — | — | — | 63.76 | — | — | — | — | |
| RLW2026.03 | 37.17 | — | — | — | — | — | 28.27 | 24.18 | 22.26 | 47.05 | 60.62 | 0.5759 | 0.241 | — | — | — | 63.77 | — | — | 7.78 | 5.67 | |
| UW2026.03 | 36.87 | — | — | — | — | — | 27.04 | 22.61 | 23.54 | 49.05 | 63.65 | 0.5446 | 0.226 | — | — | — | 63.17 | — | — | 4.05 | 3.33 | |
| Single-Task2019.11 | 27.5 | — | — | — | — | 58.9 | 17.5 | 15.2 | 34.9 | 73.3 | 85.7 | 0.62 | 0.25 | 57.9 | 85.8 | 95 | — | — | — | — | — | |
| AdaShare# Params ↓=-66.72019.11 | — | 8.8 | 7.9 | 10.1 | 8.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Cross-Stitch# Params ↓=0.02019.11 | — | -4.9 | 4.2 | 4.7 | 1.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DEN# Params ↓=-62.72019.11 | — | -9.9 | 1.7 | -35.2 | -14.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MTAN# Params ↓=+3.72019.11 | — | -4.2 | 8.7 | 3.8 | 2.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Multi-Task# Params ↓=-66.72019.11 | — | -7.6 | 7.5 | 5.2 | 1.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| NDDR-CNN# Params ↓=+5.02019.11 | — | -15 | 2.9 | -3.5 | -5.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Sluice# Params ↓=0.02019.11 | — | -8.4 | 2.9 | 4.1 | -0.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |