Method Details
Details for method 'PANet [COCO]'
Method overview
| name | PANet [COCO] |
| challenge | instance-level semantic labeling |
| details | PANet, ResNet-50 as base model, Cityscapes fine-only + COCO, training hyper-parameters are adopted from Mask R-CNN. |
| publication | Path Aggregation Network for Instance Segmentation Shu Liu, Lu Qi, Haifang Qin, Jianping Shi, Jiaya Jia CVPR 2018 https://arxiv.org/abs/1803.01534 |
| project page / code | https://github.com/ShuLiu1993/PANet |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | November, 2017 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| AP | 36.4357 |
| AP50% | 63.1375 |
| AP100m | 49.2318 |
| AP50m | 51.7758 |
Class results
| Class | AP | AP50% | AP100m | AP50m |
|---|---|---|---|---|
| person | 41.4881 | 74.4337 | 58.7145 | 58.6554 |
| rider | 33.5787 | 71.4682 | 46.6825 | 46.8548 |
| car | 58.1983 | 83.2865 | 75.7659 | 77.8564 |
| truck | 31.8024 | 42.9845 | 41.7418 | 47.6237 |
| bus | 45.348 | 65.189 | 61.8804 | 70.7855 |
| train | 28.7418 | 50.85 | 38.9328 | 42.2652 |
| motorcycle | 28.2249 | 59.6111 | 35.6902 | 35.9772 |
| bicycle | 24.1033 | 57.2767 | 34.4467 | 34.1884 |