Method Details
Details for method 'RPNet'
Method overview
| name | RPNet |
| challenge | pixel-level semantic labeling |
| details | we put forward a method for single-shot segmentation in a feature residual pyramid network (RPNet), which learns the main and residuals of segmentation by decomposing the label at different levels of residual blocks. |
| publication | Residual Pyramid Learning for Single-Shot Semantic Segmentation Xiaoyu Chen, Xiaotian Lou, Lianfa Bai, Jing Han arXiv https://arxiv.org/abs/1903.09746 |
| project page / code | https://github.com/superlxt/RPnet-Pytorch |
| used Cityscapes data | fine annotations |
| used external data | |
| runtime | 0.008 s CPU:i9-7920X GPU:NVIDIA GTX1080Ti |
| subsampling | 2 |
| submission date | December, 2018 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 68.2757 |
| iIoU Classes | 43.6211 |
| IoU Categories | 86.8322 |
| iIoU Categories | 72.3462 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.9029 | - |
| sidewalk | 81.2005 | - |
| building | 89.7614 | - |
| wall | 40.1751 | - |
| fence | 45.69 | - |
| pole | 56.3213 | - |
| traffic light | 61.6075 | - |
| traffic sign | 67.8302 | - |
| vegetation | 91.6775 | - |
| terrain | 67.9916 | - |
| sky | 94.5382 | - |
| person | 78.1914 | 57.6696 |
| rider | 57.373 | 34.1672 |
| car | 92.8674 | 87.1653 |
| truck | 48.304 | 24.5006 |
| bus | 57.8322 | 34.0465 |
| train | 56.1163 | 28.6721 |
| motorcycle | 49.6339 | 30.9058 |
| bicycle | 62.2249 | 51.8414 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.2309 | - |
| nature | 91.2929 | - |
| object | 63.1984 | - |
| sky | 94.5382 | - |
| construction | 90.2111 | - |
| human | 78.6147 | 58.9711 |
| vehicle | 91.7392 | 85.7212 |