Method Details
Details for method 'GridNet'
Method overview
| name | GridNet |
| challenge | pixel-level semantic labeling |
| details | We used a new architecture for semantic image segmentation called GridNet, following a grid pattern allowing multiple interconnected streams to work at different resolutions (see paper). We used only the training set without extra coarse annotated data (only 2975 images) and no pre-training (ImageNet) nor pre or post-processing. |
| publication | Residual Conv-Deconv Grid Network for Semantic Segmentation Damien Fourure, Rémi Emonet, Elisa Fromont, Damien Muselet, Alain Tremeau & Christian Wolf BMVC 2017 https://arxiv.org/abs/1707.07958 |
| project page / code | https://github.com/Fourure/GridNet |
| used Cityscapes data | fine annotations |
| used external data | |
| runtime | n/a |
| subsampling | no |
| submission date | August, 2017 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 69.7986 |
| iIoU Classes | 44.4807 |
| IoU Categories | 88.0848 |
| iIoU Categories | 71.4447 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.057 | - |
| sidewalk | 83.0477 | - |
| building | 90.873 | - |
| wall | 41.4479 | - |
| fence | 49.1888 | - |
| pole | 60.0824 | - |
| traffic light | 66.4862 | - |
| traffic sign | 70.1901 | - |
| vegetation | 92.4849 | - |
| terrain | 69.8092 | - |
| sky | 93.8179 | - |
| person | 82.2651 | 57.6966 |
| rider | 63.1764 | 37.1262 |
| car | 93.247 | 85.8755 |
| truck | 42.5902 | 21.9884 |
| bus | 55.7878 | 38.8128 |
| train | 48.4723 | 29.1547 |
| motorcycle | 55.3897 | 31.9926 |
| bicycle | 69.7601 | 53.199 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.4348 | - |
| nature | 92.1415 | - |
| object | 66.2026 | - |
| sky | 93.8179 | - |
| construction | 91.0514 | - |
| human | 82.3257 | 58.6702 |
| vehicle | 92.6198 | 84.2192 |