Standing dead tree segmentation on Polish dataset (unbalanced evaluation)
11Mean Tree IoUBasic KD
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Basic KDTraining Strategy=Knowledge Distillation2026.06 | 11 | 50 | 80 | 62 | 2.768 | 1,482 | 1,479 | 365 | |
| Feature-level KDTraining Strategy=Knowledge Distillation2026.06 | 10.6 | 55 | 75 | 63 | 3.039 | 1,384 | 1,134 | 463 | |
| Fine-tunedTraining Strategy=Fine-tuning2026.06 | 10.5 | 46 | 80 | 58 | 2.964 | 1,479 | 1,765 | 368 | |
| Ensemble KDTraining Strategy=Knowledge Distillation2026.06 | 8.9 | 45 | 73 | 56 | 3.719 | 1,353 | 1,647 | 494 | |
| Self KDTraining Strategy=Knowledge Distillation2026.06 | 7.4 | 28 | 82 | 42 | 4.635 | 1,524 | 3,868 | 323 |