Details for method 'RGB-D FCN'
Method overview
| name |
RGB-D FCN |
| challenge |
pixel-level semantic labeling |
| details |
GoogLeNet + depth branch, single model
no data augmentation, no training on validation set, no graphical model
Used coarse labels to initialize depth branch |
| publication |
Anonymous
|
| project page / code |
|
| used Cityscapes data |
fine annotations, coarse annotations, stereo |
| used external data |
ImageNet |
| runtime |
n/a |
| subsampling |
no |
| submission date |
January, 2017 |
| previous submissions |
|
Average results
| Metric |
Value |
| IoU Classes | 67.4204 |
| iIoU Classes | 42.1229 |
| IoU Categories | 87.4826 |
| iIoU Categories | 70.9564 |
Class results
| Class |
IoU |
iIoU |
| road | 97.9107 | - |
| sidewalk | 81.1994 | - |
| building | 90.6619 | - |
| wall | 41 | - |
| fence | 44.7835 | - |
| pole | 56.8196 | - |
| traffic light | 65.2701 | - |
| traffic sign | 69.4114 | - |
| vegetation | 91.8973 | - |
| terrain | 68.7004 | - |
| sky | 94.7467 | - |
| person | 78.924 | 55.9927 |
| rider | 52.8694 | 35.4758 |
| car | 93.1197 | 85.7791 |
| truck | 38.7713 | 22.2602 |
| bus | 53.1 | 33.495 |
| train | 43.7236 | 23.3357 |
| motorcycle | 51.0407 | 30.0223 |
| bicycle | 67.0382 | 50.6219 |
Category results
| Category |
IoU |
iIoU |
| flat | 98.3547 | - |
| nature | 91.5541 | - |
| object | 64.2866 | - |
| sky | 94.7467 | - |
| construction | 90.8811 | - |
| human | 80.2521 | 58.0457 |
| vehicle | 92.3032 | 83.8671 |
Links
Download results as .csv file
Benchmark page
