Method Details
Details for method 'Mobilenetv3-small-backbone real-time segmentation'
Method overview
| name | Mobilenetv3-small-backbone real-time segmentation |
| challenge | pixel-level semantic labeling |
| details | The model is a dual-path network with mobilenetv3-small backbone. PSP module was used as the context aggregation block. We also use feature fusion module at x16, x32. The features of the two branches are then concatenated and fused with a bottleneck conv. Only train data is used to train the model excluding validation data. And evaluation was done by single scale input images. |
| publication | Anonymous |
| project page / code | https://github.com/Chris10M/mobilenetv3-small-rt-segmentation |
| used Cityscapes data | fine annotations |
| used external data | |
| runtime | 0.02 s RTX2070 |
| subsampling | no |
| submission date | April, 2021 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 63.8835 |
| iIoU Classes | 37.7915 |
| IoU Categories | 84.349 |
| iIoU Categories | 67.4893 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.0199 | - |
| sidewalk | 75.1979 | - |
| building | 88.6545 | - |
| wall | 41.1023 | - |
| fence | 44.1955 | - |
| pole | 45.0408 | - |
| traffic light | 53.1977 | - |
| traffic sign | 61.2709 | - |
| vegetation | 90.8994 | - |
| terrain | 68.3601 | - |
| sky | 93.172 | - |
| person | 74.4238 | 52.1445 |
| rider | 50.9767 | 29.0778 |
| car | 92.4485 | 83.8811 |
| truck | 44.6049 | 16.9623 |
| bus | 46.3986 | 27.9013 |
| train | 39.6076 | 20.1281 |
| motorcycle | 45.2923 | 23.0655 |
| bicycle | 61.924 | 49.1716 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.1405 | - |
| nature | 90.5565 | - |
| object | 53.0934 | - |
| sky | 93.172 | - |
| construction | 88.9307 | - |
| human | 75.7703 | 53.5354 |
| vehicle | 90.7796 | 81.4433 |