Method Details
Details for method 'ShuffleNet v2 + DPC'
Method overview
| name | ShuffleNet v2 + DPC |
| challenge | pixel-level semantic labeling |
| details | ShuffleNet v2 with DPC at output_stride 16. |
| publication | An efficient solution for semantic segmentation: ShuffleNet V2 with atrous separable convolutions Sercan Turkmen, Janne Heikkila https://arxiv.org/abs/1902.07476 |
| project page / code | https://github.com/sercant/mobile-segmentation |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | ImageNet, MS COCO 2017 |
| runtime | n/a |
| subsampling | no |
| submission date | February, 2019 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 70.3332 |
| iIoU Classes | 43.5792 |
| IoU Categories | 86.4752 |
| iIoU Categories | 69.9179 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.1109 | - |
| sidewalk | 82.4585 | - |
| building | 90.7032 | - |
| wall | 51.3065 | - |
| fence | 50.9301 | - |
| pole | 51.4732 | - |
| traffic light | 61.2217 | - |
| traffic sign | 66.9256 | - |
| vegetation | 91.7018 | - |
| terrain | 68.5318 | - |
| sky | 93.8697 | - |
| person | 78.4704 | 55.4565 |
| rider | 59.7277 | 34.9161 |
| car | 93.9507 | 85.6976 |
| truck | 59.0531 | 26.9855 |
| bus | 68.0789 | 35.6205 |
| train | 48.0689 | 30.3736 |
| motorcycle | 54.2689 | 30.6279 |
| bicycle | 67.4792 | 48.9562 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.2582 | - |
| nature | 91.2845 | - |
| object | 59.5333 | - |
| sky | 93.8697 | - |
| construction | 90.8674 | - |
| human | 78.5613 | 56.7018 |
| vehicle | 92.9521 | 83.134 |