Details for method 'Margin calibration'
Method overview
| name |
Margin calibration |
| challenge |
pixel-level semantic labeling |
| details |
We designed a novel learning objective, distribution-aware margin calibration, for semantic segmentation. The segmentation model is DeepLab v3+ backend on SEResNeXt50. |
| publication |
Anonymous
|
| project page / code |
|
| used Cityscapes data |
fine annotations |
| used external data |
|
| runtime |
n/a |
| subsampling |
3 |
| submission date |
November, 2020 |
| previous submissions |
|
Average results
| Metric |
Value |
| IoU Classes | 81.5045 |
| iIoU Classes | 59.9271 |
| IoU Categories | 91.855 |
| iIoU Categories | 81.1092 |
Class results
| Class |
IoU |
iIoU |
| road | 98.81 | - |
| sidewalk | 87.6584 | - |
| building | 93.9779 | - |
| wall | 63.0827 | - |
| fence | 64.1883 | - |
| pole | 70.969 | - |
| traffic light | 77.4691 | - |
| traffic sign | 81.1846 | - |
| vegetation | 94.0547 | - |
| terrain | 73.0041 | - |
| sky | 95.9705 | - |
| person | 87.6073 | 71.5445 |
| rider | 71.9356 | 49.7851 |
| car | 96.4633 | 91.7298 |
| truck | 75.9031 | 44.442 |
| bus | 86.4259 | 57.0285 |
| train | 83.9141 | 53.4394 |
| motorcycle | 68.5813 | 47.3885 |
| bicycle | 77.3855 | 64.0592 |
Category results
| Category |
IoU |
iIoU |
| flat | 98.7985 | - |
| nature | 93.7397 | - |
| object | 76.5282 | - |
| sky | 95.9705 | - |
| construction | 94.2155 | - |
| human | 87.7552 | 72.3879 |
| vehicle | 95.9776 | 89.8305 |
Links
Download results as .csv file
Benchmark page
