Method Details
Details for method 'COPS'
Method overview
| name | COPS |
| challenge | panoptic semantic labeling |
| details | COPS fully differentiable with ResNet 50 backbone. |
| publication | Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach Ahmed Abbas, Paul Swoboda NeurIPS 2021 https://arxiv.org/abs/2106.03188 |
| project page / code | |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | 4 |
| submission date | October, 2021 |
| previous submissions |
Average results
| Metric | All | Things | Stuff |
|---|---|---|---|
| PQ | 59.9838 | 51.82 | 65.9211 |
| SQ | 81.4437 | 80.1981 | 82.3496 |
| RQ | 72.6364 | 64.5763 | 78.4984 |
Class results
| Class | PQ | SQ | RQ |
|---|---|---|---|
| road | 98.0901 | 98.2521 | 99.8351 |
| sidewalk | 75.0881 | 84.7179 | 88.6331 |
| building | 87.5902 | 90.1238 | 97.1888 |
| wall | 34.9552 | 76.6325 | 45.614 |
| fence | 37.0518 | 75.0747 | 49.3532 |
| pole | 51.3207 | 66.9932 | 76.6058 |
| traffic light | 53.4923 | 74.7811 | 71.5319 |
| traffic sign | 66.1987 | 77.2892 | 85.6506 |
| vegetation | 89.7999 | 91.2302 | 98.4322 |
| terrain | 42.7254 | 78.2961 | 54.569 |
| sky | 88.8195 | 92.4546 | 96.0683 |
| person | 51.8347 | 76.5209 | 67.7393 |
| rider | 49.0048 | 74.1346 | 66.1024 |
| car | 64.0499 | 83.5842 | 76.6293 |
| truck | 45.7501 | 87.5948 | 52.2293 |
| bus | 57.2395 | 89.0088 | 64.3077 |
| train | 59.593 | 83.7675 | 71.1409 |
| motorcycle | 45.3173 | 75.3199 | 60.1665 |
| bicycle | 41.7708 | 71.6546 | 58.2946 |