Semantic Segmentation on NYU v2 (mIoU, Pixel Acc, Delta Metrics)
56.93mIoUCORE-MTL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CORE-MTL2026.06 | 56.93 | 77.59 | — | — | |
| RepMTL2026.06 | 54.92 | 75.98 | — | — | |
| MTAN2026.06 | 54.01 | 75.42 | — | — | |
| STCH2026.06 | 53.77 | 74.83 | — | — | |
| Equal Weighting2026.06 | 53.16 | 75.08 | — | — | |
| MOML2026.06 | 53.03 | 74.96 | — | — | |
| RLW2026.06 | 52.93 | 75.1 | — | — | |
| FairGrad2026.06 | 52.91 | 74.27 | — | — | |
| GradNorm2026.06 | 52.64 | 74.36 | — | — | |
| PCGrad2026.06 | 52.22 | 74.33 | — | — | |
| IndividualBackbone=ResNet-502024.12 | 52 | 74.2 | — | — | |
| Single Task2026.06 | 51.92 | 73.96 | — | — | |
| SyMergeBackbone=ResNet-502024.12 | 49.8 | 73.1 | — | — | |
| ExcessMTL2026.06 | 48.46 | 71.8 | — | — | |
| ProbSurgery w/ TABackbone=ResNet-502024.12 | 43.6 | 67.6 | — | — | |
| Surgery w/ TABackbone=ResNet-502024.12 | 43.3 | 67.4 | — | — | |
| EMR-MergingBackbone=ResNet-502024.12 | 41.5 | 67.2 | — | — | |
| Ties-MergingBackbone=ResNet-502024.12 | 39.9 | 62.7 | — | — | |
| Weight AveragingBackbone=ResNet-502024.12 | 36.6 | 64 | — | — | |
| LiNeS w/ TABackbone=ResNet-502024.12 | 36.2 | 64 | — | — | |
| Task ArithmeticBackbone=ResNet-502024.12 | 31.6 | 60.3 | — | — | |
| AdaShare# Params (%)=- 50.02019.11 | 29.6 | 61.3 | 5.6 | 5.9 | |
| MTAN# Params (%)=+ 23.52019.11 | 29.5 | 60.8 | 5 | 5.4 | |
| NDDR-CNN# Params (%)=+ 6.52019.11 | 28.2 | 60.1 | 2.1 | 3.1 | |
| Single-Task# Params (%)=0.02019.11 | 27.8 | 58.5 | 0 | 0 | |
| Sluice Networks# Params (%)=0.02019.11 | 26.6 | 59.1 | 1.6 | 2.2 | |
| DEN# Params (%)=- 39.02019.11 | 26.3 | 58.8 | 2.4 | 0.6 | |
| Cross-Stitch Networks# Params (%)=0.02019.11 | 25.3 | 57.4 | 5.4 | 0.1 | |
| MagMaxBackbone=ResNet-502024.12 | 24.7 | 54.7 | — | — | |
| Multi-Task# Params (%)=- 50.02019.11 | 22.6 | 55 | 12.3 | 4.6 |