Semantic Segmentation on MarsScapes 45 (5-1 setting)
57.2IoU (Soil)Centralized (upper bound)
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Centralized (upper bound)Training Protocol=Centralized, Segmentation Backbone=DeepLabV3+2026.04 | 57.2 | 53.8 | 41.6 | 41.5 | 28.4 | 25.6 | 14.8 | 15.2 | 15.3 | 32.6 | — | |
| EIRTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 55.2 | 43.5 | 37.3 | 31.4 | 24.3 | 16.4 | 9.6 | 7.1 | 11.8 | 26.3 | 4.1 | |
| OursTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 55 | 51.8 | 39.2 | 39 | 26 | 23.7 | 12.4 | 13 | 13.4 | 30.4 | — | |
| FBLTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 53.6 | 45.2 | 39.5 | 33.2 | 25.1 | 18.5 | 11.1 | 8.4 | 13.6 | 27.6 | 2.8 | |
| ADAPTTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 51.8 | 42.2 | 35.5 | 30.1 | 26.2 | 14.6 | 8.8 | 5.8 | 10.2 | 25 | 5.4 | |
| CoMBOTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 50.1 | 40.6 | 33.9 | 28.2 | 22.4 | 13.2 | 12.5 | 4.3 | 5.4 | 23.4 | 7 | |
| CS2KTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 48.5 | 38.8 | 32.2 | 26.4 | 20.6 | 11.1 | 6.3 | 3.1 | 8.5 | 21.7 | 8.7 | |
| CUETraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 46.3 | 36.2 | 29.8 | 23.6 | 18.1 | 8.5 | 4.2 | 1.4 | 6.8 | 19.4 | 11 | |
| PLOPTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 43.2 | 33.4 | 26.6 | 19.5 | 14.2 | 5.3 | 1.8 | 0 | 4.1 | 16.5 | 13.9 | |
| Finetuning + FLTraining Protocol=Federated Learning, Segmentation Backbone=DeepLabV3+2026.04 | 32.6 | 18.4 | 0 | 0 | 0 | 0 | 0 | 0 | 4.2 | 6.1 | 24.3 |