Semantic Segmentation on Mapillary Vistas full (val)
56.9mIoU (Mean IoU)DDB
Evaluation results
| Method | Links | ||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DDBBackbone=ResNet-101, Segmentation Head=DeepLabv2, Source Dataset=GTA52022.09 | 56.9 | 89.3 | 60.8 | 81.4 | 35.9 | 38.4 | 32.9 | 48.5 | 50.5 | 69.9 | 37.9 | 90.1 | 62.6 | 49.6 | 86 | 62.7 | 62.9 | 26.1 | 52 | 42.8 | |
| Ours-EnsembleSource=S+G, Stage-wise training=true2021.03 | 53.44 | 88.4 | 40.1 | 81.9 | 32.4 | 39.8 | 41.4 | 42.2 | 42.7 | 80.1 | 46.4 | 95.6 | 58.2 | 48.5 | 84.7 | 46.6 | 45.5 | 11.7 | 46.9 | 42.4 | |
| Ours-EnsembleSource=S+G, Stage-wise training=false2021.03 | 53.37 | 88.5 | 34.3 | 81.9 | 31.9 | 41.1 | 39 | 40.1 | 41.5 | 79.7 | 45 | 95.7 | 62.7 | 51.1 | 83.3 | 49.9 | 45.9 | 8.5 | 46.4 | 47.5 | |
| Ours-M2Source=S+G, Stage-wise training=true2021.03 | 52.37 | 88.6 | 36.5 | 81.4 | 29.7 | 38.2 | 41.3 | 43 | 43.4 | 80.2 | 45.8 | 95.6 | 58.3 | 43.8 | 84.5 | 42.5 | 42 | 10.1 | 46.2 | 43.9 | |
| Ours-M1Source=S+G, Stage-wise training=true2021.03 | 51.92 | 87.5 | 40.1 | 80.9 | 31 | 37.4 | 40 | 42.5 | 40.6 | 79.6 | 42.4 | 95.2 | 55.5 | 46.5 | 84.5 | 45.1 | 40.3 | 16.5 | 41.6 | 39.1 | |
| Ours-M2Source=S+G, Stage-wise training=false2021.03 | 51.67 | 87.8 | 31.6 | 81 | 30 | 37.8 | 34.8 | 38.3 | 41.3 | 78.1 | 39.1 | 95.1 | 60.1 | 49.5 | 82.2 | 42.7 | 39 | 19.2 | 45.9 | 48 | |
| Ours-M1Source=S+G, Stage-wise training=false2021.03 | 50.89 | 88.2 | 32.5 | 81 | 29.1 | 37.5 | 39.9 | 41.7 | 39.6 | 80.4 | 44.6 | 95.8 | 58.7 | 40.2 | 83.1 | 48.1 | 40.7 | 2.3 | 40.1 | 43.2 | |
| AdventSource=S+G, Stage-wise training=false2021.03 | 49.4 | 87.2 | 36.2 | 78 | 27.1 | 31.2 | 38.4 | 40.8 | 40.2 | 80.8 | 44.2 | 96 | 47.1 | 43.5 | 82.3 | 39 | 39.3 | 5 | 42 | 40.3 | |
| ADASBackbone=ResNet-101, Segmentation Head=DeepLabv2, Source Dataset=GTA52022.09 | 49.2 | 84.2 | 33.9 | 78.5 | 25.5 | 24.5 | 35.6 | 39.8 | 52.4 | 71.2 | 40.2 | 92.4 | 58.7 | 38.7 | 82.7 | 44.4 | 46.4 | 15.2 | 37.8 | 32.2 | |
| AdaptSegSource=S+G, Stage-wise training=false2021.03 | 48.19 | 84.2 | 33.4 | 78 | 27.9 | 34 | 38 | 41.6 | 39.4 | 78.6 | 34.5 | 92.7 | 46.9 | 41.6 | 81.9 | 38.3 | 39 | 3.6 | 41.5 | 40.5 | |
| AdventSource=GTA5 (G), Stage-wise training=false2021.03 | 47.79 | 86.6 | 28.3 | 77.9 | 24.7 | 30.6 | 36.1 | 36 | 32.5 | 75.8 | 34.9 | 94.4 | 58.8 | 44.1 | 79.9 | 41.3 | 42.3 | 15.7 | 35.6 | 32.6 | |
| AdaptSegSource=GTA5 (G), Stage-wise training=false2021.03 | 47.44 | 86.5 | 31.6 | 78.2 | 24.6 | 30 | 36.1 | 35.8 | 31.6 | 73.4 | 33.2 | 93.7 | 59.2 | 44.5 | 78.6 | 41.2 | 39.3 | 14.8 | 36.5 | 32.3 | |
| CCLBackbone=ResNet-101, Segmentation Head=DeepLabv2, Source Dataset=GTA52022.09 | 46.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DTSource=S+G, Stage-wise training=false2021.03 | 46.64 | 77.7 | 30.9 | 75.2 | 27 | 27.5 | 33.4 | 37.2 | 37.3 | 76.9 | 43.1 | 93.3 | 55.8 | 38 | 72.5 | 38.4 | 40.2 | 2.8 | 36.9 | 42.3 | |
| DTSource=GTA5 (G), Stage-wise training=false2021.03 | 42.89 | 82.2 | 28.6 | 74.2 | 23.4 | 27.2 | 35.3 | 36.4 | 18.6 | 73.8 | 29.2 | 89.6 | 58.9 | 39.2 | 74.5 | 35 | 17.2 | 12.5 | 31.3 | 27.8 | |
| AdventSource=Synscapes (S), Stage-wise training=false2021.03 | 42.04 | 86.2 | 23.9 | 74.6 | 17.8 | 26.8 | 29.5 | 35.9 | 39.8 | 79.4 | 43.6 | 96.2 | 37.3 | 27.5 | 78.4 | 26.3 | 16.1 | 1.4 | 29.1 | 29.1 | |
| AdaptSegSource=Synscapes (S), Stage-wise training=false2021.03 | 41.35 | 85.9 | 24.2 | 73.2 | 17.7 | 27.4 | 26.4 | 33 | 39 | 75.4 | 44.6 | 94.3 | 34.7 | 27.8 | 77.4 | 25.8 | 16.5 | 1.2 | 29.9 | 31.2 | |
| DTSource=Synscapes (S), Stage-wise training=false2021.03 | 37.18 | 70.4 | 23.6 | 63.6 | 14.8 | 12 | 25.8 | 30.7 | 32.7 | 75.2 | 41.2 | 89.4 | 36.2 | 22 | 73 | 19.5 | 17.2 | 0.2 | 27.7 | 31.1 | |
| Source onlyBackbone=ResNet-101, Segmentation Head=DeepLabv2, Source Dataset=GTA52022.09 | 35.5 | 55.7 | 27.1 | 55.3 | 9.9 | 20.6 | 22.7 | 33.3 | 31.6 | 68.4 | 21.1 | 70.6 | 53.5 | 30.9 | 72.7 | 32.3 | 11.6 | 5.6 | 36.3 | 14.9 |