Method Details
Details for method 'LiteSeg-Mobilenet'
Method overview
| name | LiteSeg-Mobilenet |
| challenge | pixel-level semantic labeling |
| details | |
| publication | LiteSeg: A Litewiegth ConvNet for Semantic Segmentation Taha Emara, Hossam E. Abd El Munim, Hazem M. Abbas DICTA 2019 https://arxiv.org/abs/1912.06683 |
| project page / code | https://github.com/tahaemara/LiteSeg |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | |
| runtime | 0.0062 s Intel Core i7-8700 @ 3.2GHZ, 16GB memory, and NVIDIA GTX1080Ti GPU card. Input image resolution 360X640. |
| subsampling | no |
| submission date | January, 2019 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 67.8178 |
| iIoU Classes | 45.2536 |
| IoU Categories | 86.7996 |
| iIoU Categories | 72.0442 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.066 | - |
| sidewalk | 77.7848 | - |
| building | 90.4489 | - |
| wall | 44.1649 | - |
| fence | 47.4571 | - |
| pole | 55.3224 | - |
| traffic light | 59.2663 | - |
| traffic sign | 68.6587 | - |
| vegetation | 91.978 | - |
| terrain | 69.1893 | - |
| sky | 94.6277 | - |
| person | 78.6432 | 60.5537 |
| rider | 55.2693 | 37.2237 |
| car | 92.0265 | 83.9742 |
| truck | 42.9246 | 26.3897 |
| bus | 55.8204 | 35.0045 |
| train | 54.7704 | 33.7944 |
| motorcycle | 49.165 | 33.1627 |
| bicycle | 63.9557 | 51.9262 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 97.9082 | - |
| nature | 91.7048 | - |
| object | 62.7535 | - |
| sky | 94.6277 | - |
| construction | 90.446 | - |
| human | 79.2692 | 62.0914 |
| vehicle | 90.8878 | 81.997 |