Method Details
Details for method 'ResNet-38'
Method overview
| name | ResNet-38 |
| challenge | pixel-level semantic labeling |
| details | single model, no post-processing with CRFs Model A2, 2 conv., fine+coarse, multi scale testing |
| publication | Wider or Deeper: Revisiting the ResNet Model for Visual Recognition Zifeng Wu, Chunhua Shen, Anton van den Hengel arxiv https://arxiv.org/abs/1611.10080 |
| project page / code | https://github.com/itijyou/ademxapp |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | ImageNet, Places |
| runtime | n/a |
| subsampling | no |
| submission date | December, 2016 |
| previous submissions | 1 |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 80.6055 |
| iIoU Classes | 57.7697 |
| IoU Categories | 91.038 |
| iIoU Categories | 79.0781 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.6821 | - |
| sidewalk | 86.9483 | - |
| building | 93.3276 | - |
| wall | 60.4011 | - |
| fence | 62.8502 | - |
| pole | 67.5784 | - |
| traffic light | 74.956 | - |
| traffic sign | 78.6961 | - |
| vegetation | 93.7236 | - |
| terrain | 73.6716 | - |
| sky | 95.4527 | - |
| person | 86.7896 | 68.5126 |
| rider | 71.1327 | 48.7641 |
| car | 96.0795 | 90.487 |
| truck | 75.1981 | 42.0183 |
| bus | 87.5849 | 51.8896 |
| train | 81.8735 | 50.6861 |
| motorcycle | 69.8367 | 46.974 |
| bicycle | 76.7225 | 62.8255 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.7375 | - |
| nature | 93.3941 | - |
| object | 73.588 | - |
| sky | 95.4527 | - |
| construction | 93.6488 | - |
| human | 86.9407 | 69.6379 |
| vehicle | 95.5043 | 88.5182 |