Multi-task learning on Tiny-Taskonomy (test)
-28.2Delta t1 (Semantic Segmentation)DEN
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DEN# Params (Rel. %)=-77.62021.10 | -28.2 | -2.6 | -22.7 | 2.5 | 4.2 | -9.3 | |
| MTAN# Params (Rel. %)=-9.82021.10 | -8 | -2.5 | -4.5 | 0 | 2.8 | -2.4 | |
| NDDR# Params (Rel. %)=+8.22021.10 | -4.2 | -0.9 | -4.5 | 0.5 | 4.2 | -1 | |
| Multi-Task# Params (Rel. %)=-80.02021.10 | -3.7 | -1.4 | -4.5 | 0 | 4.2 | -1.1 | |
| Sluice# Params (Rel. %)=+0.02021.10 | -3.7 | -1.5 | -9.1 | 0.5 | 2.4 | -2.3 | |
| Cross-Stitch# Params (Rel. %)=+0.02021.10 | 0.9 | -3.5 | 0 | -1 | -2.4 | -1.2 | |
| AdaShare# Params (Rel. %)=-80.02021.10 | 2.3 | -0.6 | -4.5 | 3 | 5.7 | 1.2 | |
| AutoMTL# Params (Rel. %)=-50.12021.10 | 3 | 8.2 | 0 | 3 | 7.1 | 4.3 | |
| Learn to Branch# Params (Rel. %)=-71.22021.10 | 9.4 | 5.3 | -4.5 | -2.5 | -2.4 | 1.1 |