Method Details
Details for method 'PSPNet'
Method overview
| name | PSPNet |
| challenge | pixel-level semantic labeling |
| details | This submission is trained on coarse+fine(train+val set, 2975+500 images). Former submission is trained on coarse+fine(train set, 2975 images) which gets 80.2 mIoU: https://www.cityscapes-dataset.com/method-details/?submissionID=314 Previous versions of this method were listed as "SenseSeg_1026". |
| publication | Pyramid Scene Parsing Network Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, Jiaya Jia CVPR 2017 https://arxiv.org/abs/1612.01105 |
| project page / code | https://hszhao.github.io/projects/pspnet |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | May, 2017 |
| previous submissions | 1, 2, 3, 4, 5, 6 |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 81.1947 |
| iIoU Classes | 59.6145 |
| IoU Categories | 91.15 |
| iIoU Categories | 79.2468 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.6814 | - |
| sidewalk | 86.9233 | - |
| building | 93.4688 | - |
| wall | 58.3855 | - |
| fence | 63.6752 | - |
| pole | 67.6716 | - |
| traffic light | 76.1225 | - |
| traffic sign | 80.474 | - |
| vegetation | 93.6399 | - |
| terrain | 72.2031 | - |
| sky | 95.2983 | - |
| person | 86.8317 | 69.3269 |
| rider | 71.9132 | 51.1528 |
| car | 96.2174 | 90.2703 |
| truck | 77.6988 | 42.638 |
| bus | 91.5102 | 55.1206 |
| train | 83.6401 | 56.2058 |
| motorcycle | 70.8048 | 48.6919 |
| bicycle | 77.5388 | 63.5096 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.7071 | - |
| nature | 93.3461 | - |
| object | 74.192 | - |
| sky | 95.2983 | - |
| construction | 93.7622 | - |
| human | 87.0754 | 70.2479 |
| vehicle | 95.6688 | 88.2457 |
