Method Details
Details for method 'UPSNet'
Method overview
| name | UPSNet |
| challenge | instance-level semantic labeling |
| details | |
| publication | UPSNet: A Unified Panoptic Segmentation Network Yuwen Xiong, Renjie Liao, Hengshuang Zhao, Rui Hu, Min Bai, Ersin Yumer, Raquel Urtasun CVPR 2019 https://arxiv.org/abs/1901.03784 |
| project page / code | https://github.com/uber-research/UPSNet |
| used Cityscapes data | fine annotations |
| used external data | ImageNet, COCO |
| runtime | 0.227 s 1080 Ti |
| subsampling | no |
| submission date | March, 2019 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| AP | 33.039 |
| AP50% | 59.648 |
| AP100m | 46.8193 |
| AP50m | 50.7484 |
Class results
| Class | AP | AP50% | AP100m | AP50m |
|---|---|---|---|---|
| person | 35.9429 | 69.1163 | 52.6477 | 52.8816 |
| rider | 27.4023 | 66.4228 | 39.9304 | 40.6705 |
| car | 51.8544 | 76.8932 | 71.0664 | 73.6474 |
| truck | 31.7785 | 44.6222 | 44.4484 | 52.4584 |
| bus | 43.0884 | 62.435 | 62.0179 | 72.9218 |
| train | 31.3841 | 54.6125 | 44.9886 | 53.2922 |
| motorcycle | 23.7622 | 53.8755 | 30.9894 | 31.6711 |
| bicycle | 19.0994 | 49.2062 | 28.4658 | 28.4443 |