Method Details
Details for method 'HRNetV2 + OCR + SegFix'
Method overview
| name | HRNetV2 + OCR + SegFix |
| challenge | pixel-level semantic labeling |
| details | First, we pre-train "HRNet+OCR" method on the Mapillary training set (achieves 50.8% on the Mapillary val set). Second, we fine-tune the model with the Cityscapes training, validation and coarse set. Finally, we apply the "SegFix" scheme to further improve the results. |
| publication | Object-Contextual Representations for Semantic Segmentation Yuhui Yuan, Xilin Chen, Jingdong Wang https://arxiv.org/abs/1909.11065 |
| project page / code | https://github.com/openseg-group/openseg.pytorch |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | ImageNet, Mapillary |
| runtime | n/a |
| subsampling | no |
| submission date | January, 2020 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 84.5008 |
| iIoU Classes | 65.9364 |
| IoU Categories | 92.6646 |
| iIoU Categories | 83.8776 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.8884 | - |
| sidewalk | 88.3393 | - |
| building | 94.3989 | - |
| wall | 67.9743 | - |
| fence | 67.8259 | - |
| pole | 73.597 | - |
| traffic light | 80.6042 | - |
| traffic sign | 83.9262 | - |
| vegetation | 94.35 | - |
| terrain | 74.4519 | - |
| sky | 96.0615 | - |
| person | 89.2148 | 76.2622 |
| rider | 75.8517 | 56.9215 |
| car | 96.8302 | 91.9276 |
| truck | 83.6267 | 51.2988 |
| bus | 94.1788 | 65.179 |
| train | 91.2842 | 62.6755 |
| motorcycle | 74.0213 | 54.9142 |
| bicycle | 80.09 | 68.3127 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.8475 | - |
| nature | 94.0198 | - |
| object | 79.1512 | - |
| sky | 96.0615 | - |
| construction | 94.6439 | - |
| human | 89.4301 | 77.3843 |
| vehicle | 96.4981 | 90.3708 |