Multi-task Dense Prediction on PASCAL-Context
84.67ODS F-ScorePRISM
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PRISMBackbone=ViT-L, Pre-training=ImageNet-1K, Evaluation Protocol=fine-tuning2026.06 | 84.67 | 84.34 | 77.83 | 13.43 | — | — | 3.16 | |
| SAKBackbone=ViT-L, Pre-training=ImageNet-1K, Evaluation Protocol=fine-tuning2026.06 | 84.65 | 84.01 | 76.99 | 13.82 | — | — | 2.3 | |
| Single-task baselineBackbone=ViT-L, Pre-training=ImageNet-1K, Evaluation Protocol=fine-tuning2026.06 | 83.8 | 81.61 | 72.77 | 13.87 | — | — | 0 | |
| single task baselineBackbone=Swin-T, Params (M)=140.8, Dec (%)=2.32023.07 | 76.51 | 67.7 | 66.14 | 13.37 | 65.33 | 0 | — | |
| multi-task baselineBackbone=Swin-T, Params (M)=30.8, Dec (%)=10.62023.07 | 74.68 | 65.78 | 60.68 | 13.95 | 64.44 | -3.84 | — | |
| TSNBackbone=Swin-T, Params (M)=39.1, Dec (%)=29.62023.07 | 74.04 | 67.3 | 61.11 | 14.55 | 64.29 | -4.37 | — | |
| PGTBackbone=Swin-T, Params (M)=28.5, Dec (%)=2.72023.07 | 73.93 | 67.58 | 62.58 | 13.95 | 65.59 | -2.57 | — | |
| ASTMTBackbone=ResNet-50, Params (M)=49.4, Dec (%)=13.62023.07 | 72.4 | 68 | 61.12 | 14.68 | 65.71 | -4.35 | — | |
| TSNBackbone=ResNet-34, Params (M)=28.4, Dec (%)=25.02023.07 | 71.8 | 67.6 | 58 | 16.1 | 64.3 | -8.12 | — | |
| RCMBackbone=ResNet-18, Params (M)=46.1, Dec (%)=58.92023.07 | 71.3 | 65.7 | 58.12 | 13.7 | 66.38 | -4.55 | — |