Method Details
Details for method 'SwiftNetRN-18'
Method overview
| name | SwiftNetRN-18 |
| challenge | pixel-level semantic labeling |
| details | |
| publication | In Defense of Pre-trained ImageNet Architectures for Real-time Semantic Segmentation of Road-driving Images Marin Oršić, Ivan Krešo, Petra Bevandić, Siniša Šegvić CVPR 2019 https://openaccess.thecvf.com/content_CVPR_2019/papers/Orsic_In_Defense_of_Pre-Trained_ImageNet_Architectures_for_Real-Time_Semantic_Segmentation_CVPR_2019_paper.pdf |
| project page / code | https://github.com/orsic/swiftnet |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | 0.0243 s GTX1080Ti |
| subsampling | no |
| submission date | November, 2018 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 75.5106 |
| iIoU Classes | 51.9937 |
| IoU Categories | 89.8312 |
| iIoU Categories | 77.1644 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.3159 | - |
| sidewalk | 83.8554 | - |
| building | 92.2075 | - |
| wall | 46.3184 | - |
| fence | 52.7606 | - |
| pole | 63.2448 | - |
| traffic light | 70.5667 | - |
| traffic sign | 75.8098 | - |
| vegetation | 93.1035 | - |
| terrain | 70.3177 | - |
| sky | 95.429 | - |
| person | 84.0265 | 65.4996 |
| rider | 64.5358 | 43.5209 |
| car | 95.2662 | 89.1742 |
| truck | 63.8568 | 33.0262 |
| bus | 77.9548 | 43.8256 |
| train | 71.9326 | 44.3719 |
| motorcycle | 61.5698 | 37.7429 |
| bicycle | 73.6306 | 58.7882 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.6104 | - |
| nature | 92.7508 | - |
| object | 70.0035 | - |
| sky | 95.429 | - |
| construction | 92.5975 | - |
| human | 84.7349 | 67.2306 |
| vehicle | 94.6925 | 87.0982 |