Glass Segmentation on Trans10K-Stuff
95.1IoUL+GNet
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| L+GNettraining=all datasets, binary prediction confidences=true2026.03 | 95.1 | 97.3 | 2 | 2.05 | |
| L+GNettraining mode=trained and tested separately on respective datasets, with binary prediction confidences=true2026.03 | 94.7 | 97.2 | 2.2 | 2.24 | |
| GlassWizardtraining mode=trained and tested separately on respective datasets2026.03 | 93 | 96.5 | 2.8 | 2.91 | |
| GlassWizardtraining=all datasets2026.03 | 92.8 | 96.5 | 3 | 3.04 | |
| VBNettraining mode=trained and tested separately on respective datasets2026.03 | 91.6 | 95.5 | 3.2 | 3.41 | |
| PGSNettraining mode=trained and tested separately on respective datasets2026.03 | 89.8 | — | 4.2 | 4.39 | |
| GSDNettraining mode=trained and tested separately on respective datasets2026.03 | 89.7 | — | 4.2 | 4.52 | |
| GDNettraining mode=trained and tested separately on respective datasets2026.03 | 88.7 | — | 4.6 | 4.72 | |
| TransLabtraining mode=trained and tested separately on respective datasets2026.03 | 87.1 | — | 5.1 | 5.44 | |
| Trans2Segtraining mode=trained and tested separately on respective datasets2026.03 | 75 | — | 12.4 | 10.73 |