Multi-task Learning on PASCAL-Context (test)
4.31HVSingle-Task
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Single-TaskInference Params.=5.91M, GFLOPS=9.752022.03 | 4.31 | — | — | |
| DYMUc=0.0, adaptation=true, Inference Params.=3.15M, GFLOPS=5.52, Control Params.=9.25M2022.03 | 4.26 | 0.82 | — | |
| DYMUc=1.0, adaptation=true, Inference Params.=2.86M, GFLOPS=5.07, Control Params.=9.25M2022.03 | 4.25 | 0.82 | — | |
| PHN-BNInference Params.=2.06M, GFLOPS=4.81, Control Params.=3.32M2022.03 | 3.92 | 0.79 | — | |
| DYMU (w/o adaptation)c=0.0, adaptation=false, Inference Params.=3.15M, GFLOPS=5.52, Control Params.=0.03M2022.03 | 3.56 | 0.92 | — | |
| DYMU (w/o adaptation)c=1.0, adaptation=false, Inference Params.=2.86M, GFLOPS=5.07, Control Params.=0.03M2022.03 | 3.35 | 0.91 | — | |
| PHNInference Params.=2.06M, GFLOPS=4.81, Control Params.=21.10M2022.03 | 1.97 | 0.74 | — | |
| Aligned-MTLBackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 1.83 | |
| CAGradBackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 1.14 | |
| GDBackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 1.47 | |
| IndependentBackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 0 | |
| MGDABackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 1.15 | |
| OursBackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 3.7 | |
| PCGradBackbone=HRNet-18, Task weights=manually tuned, Experimental seeds=average of 32024.06 | — | — | 1.86 |