Method Details
Details for method 'kMaX-DeepLab [Cityscapes-fine]'
Method overview
| name | kMaX-DeepLab [Cityscapes-fine] |
| challenge | pixel-level semantic labeling |
| details | kMaX-DeepLab w/ ConvNeXt-L backbone (ImageNet-22k + 1k pretrained). This result is obtained by the kMaX-DeepLab trained for Panoptic Segmentation task. No test-time augmentation or other external dataset. |
| publication | k-means Mask Transformer Qihang Yu, Huiyu Wang, Siyuan Qiao, Maxwell Collins, Yukun Zhu, Hartwig Adam, Alan Yuille, and Liang-Chieh Chen ECCV 2022 https://arxiv.org/abs/2207.04044 |
| project page / code | https://github.com/google-research/deeplab2 |
| used Cityscapes data | fine annotations |
| used external data | ImageNet 22k + 1k |
| runtime | n/a |
| subsampling | no |
| submission date | March, 2022 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 83.1551 |
| iIoU Classes | 65.8603 |
| IoU Categories | 92.3151 |
| iIoU Categories | 82.6771 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.8487 | - |
| sidewalk | 88.2571 | - |
| building | 94.2578 | - |
| wall | 63.5207 | - |
| fence | 67.5075 | - |
| pole | 71.8943 | - |
| traffic light | 79.7129 | - |
| traffic sign | 83.2563 | - |
| vegetation | 94.0891 | - |
| terrain | 73.5139 | - |
| sky | 96.1995 | - |
| person | 88.7009 | 76.0701 |
| rider | 77.529 | 58.4487 |
| car | 96.6748 | 89.4182 |
| truck | 83.1386 | 54.4387 |
| bus | 90.3213 | 66.9697 |
| train | 82.6068 | 60.4399 |
| motorcycle | 75.4474 | 58.7047 |
| bicycle | 74.471 | 62.3926 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.7857 | - |
| nature | 93.8191 | - |
| object | 77.7917 | - |
| sky | 96.1995 | - |
| construction | 94.5083 | - |
| human | 88.9364 | 76.9856 |
| vehicle | 96.1647 | 88.3687 |