Details for method 'Deep Watershed Transformation'
Method overview
| name |
Deep Watershed Transformation |
| challenge |
instance-level semantic labeling |
| details |
Exactly the method as (DWT 2016-11-06 05:23:46), with same training input, procedures, and parameters. Fixed input normalization bug causing half of the pixel intensities to have a large undesired offset due to improper type casting. |
| publication |
Anonymous
|
| project page / code |
|
| used Cityscapes data |
fine annotations |
| used external data |
ImageNet |
| runtime |
n/a |
| subsampling |
2 |
| submission date |
November, 2016 |
| previous submissions |
1 |
Average results
| Metric |
Value |
| AP | 15.6341 |
| AP50% | 30.0028 |
| AP100m | 26.2106 |
| AP50m | 31.7666 |
Class results
| Class |
AP | AP50% | AP100m | AP50m |
| person | 15.1158 | 33.8637 | 27.0197 | 27.0484 |
| rider | 11.7496 | 31.4683 | 19.7719 | 20.4162 |
| car | 32.9057 | 49.8894 | 52.8324 | 56.9658 |
| truck | 17.0735 | 25.6732 | 28.9888 | 39.6184 |
| bus | 20.4013 | 32.5705 | 36.4466 | 51.2562 |
| train | 15.0084 | 28.1717 | 25.1199 | 37.9115 |
| motorcycle | 7.94647 | 22.5719 | 11.6686 | 12.7543 |
| bicycle | 4.87196 | 15.8137 | 7.83727 | 8.16188 |
Links
Download results as .csv file
Benchmark page
