Method Details
Details for method 'EDANet'
Method overview
| name | EDANet |
| challenge | pixel-level semantic labeling |
| details | Training data: Fine annotations only (train+val. set, 2975+500 images) without any pretraining nor coarse annotations. For training on fine annotations (train set only, 2975 images), it attains a mIoU of 66.3%. Runtime: (resolution 512x1024) 0.0092s on a single GTX 1080Ti, 0.0123s on a single Titan X. |
| publication | Efficient Dense Modules of Asymmetric Convolution for Real-Time Semantic Segmentation Shao-Yuan Lo (NCTU), Hsueh-Ming Hang (NCTU), Sheng-Wei Chan (ITRI), Jing-Jhih Lin (ITRI) https://arxiv.org/abs/1809.06323 |
| project page / code | https://github.com/shaoyuanlo/EDANet |
| used Cityscapes data | fine annotations |
| used external data | |
| runtime | 0.0092 s GTX 1080Ti |
| subsampling | 2 |
| submission date | August, 2018 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 67.3152 |
| iIoU Classes | 41.7828 |
| IoU Categories | 85.7516 |
| iIoU Categories | 69.9239 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.7698 | - |
| sidewalk | 80.6318 | - |
| building | 89.5359 | - |
| wall | 41.9571 | - |
| fence | 45.9684 | - |
| pole | 52.3381 | - |
| traffic light | 59.8394 | - |
| traffic sign | 64.9868 | - |
| vegetation | 91.3691 | - |
| terrain | 68.655 | - |
| sky | 93.5875 | - |
| person | 75.7286 | 54.9287 |
| rider | 54.2667 | 32.7323 |
| car | 92.4132 | 85.1025 |
| truck | 40.8629 | 18.5793 |
| bus | 58.7039 | 35.4491 |
| train | 55.9702 | 31.9444 |
| motorcycle | 50.4192 | 28.4388 |
| bicycle | 63.985 | 47.0872 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.1298 | - |
| nature | 90.9651 | - |
| object | 59.6403 | - |
| sky | 93.5875 | - |
| construction | 89.8326 | - |
| human | 76.5412 | 56.597 |
| vehicle | 91.5649 | 83.2509 |