Method Details
Details for method 'seamseg_rvcsubset'
Method overview
| name | seamseg_rvcsubset |
| challenge | panoptic semantic labeling |
| details | Seamless Scene Segmentation Resnet101, pretrained on Imagenet; supplied with altered MVD to include WildDash2 classes; does not contain other RVC label policies (i.e. no ADE20K/COCO-specific classes -> rvcsubset and not a proper submission) |
| publication | Seamless Scene Segmentation Porzi, Lorenzo and Rota Bulò, Samuel and Colovic, Aleksander and Kontschieder, Peter The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019 https://arxiv.org/abs/1905.01220 |
| project page / code | https://github.com/mapillary/seamseg |
| used Cityscapes data | |
| used external data | model pre-trained on ImageNet; regular training on altered MVD to contain WildDash2 labels (van, pickup) (20k frames) no other dataset (i.e. no Cityscapes frames) |
| runtime | n/a |
| subsampling | no |
| submission date | August, 2020 |
| previous submissions |
Average results
| Metric | All | Things | Stuff |
|---|---|---|---|
| PQ | 51.8745 | 41.4418 | 59.4619 |
| SQ | 78.4834 | 77.965 | 78.8604 |
| RQ | 64.7833 | 53.0156 | 73.3417 |
Class results
| Class | PQ | SQ | RQ |
|---|---|---|---|
| road | 86.9908 | 88.6266 | 98.1543 |
| sidewalk | 59.7815 | 79.0859 | 75.5906 |
| building | 85.1142 | 87.4249 | 97.357 |
| wall | 30.7843 | 73.4401 | 41.9175 |
| fence | 30.1535 | 70.855 | 42.5566 |
| pole | 50.227 | 67.2387 | 74.6996 |
| traffic light | 47.0481 | 72.7107 | 64.7059 |
| traffic sign | 58.481 | 75.8803 | 77.0701 |
| vegetation | 88.7359 | 90.6246 | 97.916 |
| terrain | 29.0057 | 68.9007 | 42.0979 |
| sky | 87.7589 | 92.6767 | 94.6936 |
| person | 49.7016 | 76.5086 | 64.9621 |
| rider | 40.7097 | 72.2438 | 56.3504 |
| car | 57.9442 | 84.6515 | 68.4502 |
| truck | 39.217 | 84.8857 | 46.1997 |
| bus | 49.056 | 84.4495 | 58.0892 |
| train | 25.5634 | 79.3811 | 32.2034 |
| motorcycle | 36.8715 | 71.875 | 51.2994 |
| bicycle | 32.471 | 69.7249 | 46.5702 |