Method Details
Details for method 'BshapeNet+ [COCO]'
Method overview
| name | BshapeNet+ [COCO] |
| challenge | instance-level semantic labeling |
| details | BshapeNet+ single model, ResNet-50-FPN as base model, Cityscapes [fine-only + COCO] |
| publication | BshapeNet: Object Detection and Instance Segmentation with Bounding Shape Masks Ba Rom Kang, Ha Young Kim https://arxiv.org/abs/1810.10327 |
| project page / code | |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | May, 2019 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| AP | 32.9144 |
| AP50% | 58.8302 |
| AP100m | 47.2968 |
| AP50m | 50.6802 |
Class results
| Class | AP | AP50% | AP100m | AP50m |
|---|---|---|---|---|
| person | 36.6013 | 70.6891 | 53.6978 | 53.9908 |
| rider | 24.7989 | 63.1195 | 37.2438 | 38.1203 |
| car | 50.4212 | 75.131 | 70.2647 | 72.9001 |
| truck | 33.687 | 46.7379 | 46.0907 | 52.9655 |
| bus | 40.9596 | 57.5251 | 61.1309 | 69.43 |
| train | 33.7342 | 52.9776 | 49.0155 | 56.7799 |
| motorcycle | 25.3615 | 56.5513 | 33.5783 | 33.9094 |
| bicycle | 17.7517 | 47.9104 | 27.3529 | 27.3456 |