Method Details
Details for method 'NVSegNet'
Method overview
| name | NVSegNet |
| challenge | pixel-level semantic labeling |
| details | train on downsampled images (2x on each side), training takes 20 hours. The model in evaluation only trained on GTfine train |
| publication | Anonymous |
| project page / code | |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | 0.4 s 1 Nvidia Titan X, Intel i7 |
| subsampling | 2 |
| submission date | May, 2016 |
| previous submissions | 1 |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 64.3906 |
| iIoU Classes | 35.9145 |
| IoU Categories | 84.9152 |
| iIoU Categories | 61.9844 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.6724 | - |
| sidewalk | 79.4634 | - |
| building | 88.8435 | - |
| wall | 36.2357 | - |
| fence | 34.4941 | - |
| pole | 50.758 | - |
| traffic light | 53.9471 | - |
| traffic sign | 61.6726 | - |
| vegetation | 90.4269 | - |
| terrain | 67.0671 | - |
| sky | 94.1366 | - |
| person | 74.5201 | 44.0166 |
| rider | 50.412 | 27.0847 |
| car | 92.4053 | 80.4622 |
| truck | 40.4929 | 18.4627 |
| bus | 51.5501 | 30.2206 |
| train | 48.2813 | 22.4551 |
| motorcycle | 49.1474 | 24.5191 |
| bicycle | 61.8955 | 40.0946 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.1388 | - |
| nature | 90.1422 | - |
| object | 57.1241 | - |
| sky | 94.1366 | - |
| construction | 88.8945 | - |
| human | 75.2084 | 45.6545 |
| vehicle | 90.7617 | 78.3143 |