Method Details
Details for method 'Real-time FCN'
Method overview
| name | Real-time FCN |
| challenge | pixel-level semantic labeling |
| details | Combines the following concepts: Network architecture: "Going deeper with convolutions". Szegedy et al., CVPR 2015 Framework and skip connections: "Fully convolutional networks for semantic segmentation". Long et al., CVPR 2015 Context modules: "Multi-scale context aggregation by dilated convolutions". Yu and Kolutin, ICLR 2016 |
| publication | Understanding Cityscapes: Efficient Urban Semantic Scene Understanding Marius Cordts Dissertation http://tuprints.ulb.tu-darmstadt.de/6893/ |
| project page / code | |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | ImageNet |
| runtime | 0.044 s Nvidia Titan X (Pascal) |
| subsampling | no |
| submission date | April, 2017 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 72.6319 |
| iIoU Classes | 45.4929 |
| IoU Categories | 87.9307 |
| iIoU Categories | 71.6014 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.9647 | - |
| sidewalk | 81.3638 | - |
| building | 91.1343 | - |
| wall | 44.626 | - |
| fence | 50.6688 | - |
| pole | 57.3114 | - |
| traffic light | 64.1286 | - |
| traffic sign | 71.1938 | - |
| vegetation | 92.0659 | - |
| terrain | 68.5463 | - |
| sky | 94.6591 | - |
| person | 81.1929 | 59.1618 |
| rider | 61.1798 | 36.1063 |
| car | 94.6045 | 85.1646 |
| truck | 54.4688 | 25.5687 |
| bus | 76.4568 | 40.0096 |
| train | 72.1555 | 35.5849 |
| motorcycle | 57.6049 | 32.6937 |
| bicycle | 68.6808 | 49.6538 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.3979 | - |
| nature | 91.515 | - |
| object | 64.3136 | - |
| sky | 94.6591 | - |
| construction | 91.3526 | - |
| human | 81.5732 | 60.477 |
| vehicle | 93.7034 | 82.7259 |