Method Details
Details for method 'Qualcomm AI Research'
Method overview
| name | Qualcomm AI Research |
| challenge | pixel-level semantic labeling |
| details | |
| publication | InverseForm: A Loss Function for Structured Boundary-Aware Segmentation Shubhankar Borse, Ying Wang, Yizhe Zhang, Fatih Porikli CVPR 2021 oral https://arxiv.org/abs/2104.02745 |
| project page / code | https://github.com/Qualcomm-AI-research/InverseForm |
| used Cityscapes data | fine annotations, coarse annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | July, 2021 |
| previous submissions | 1, 2 |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 85.7748 |
| iIoU Classes | 71.9628 |
| IoU Categories | 93.1143 |
| iIoU Categories | 85.6423 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.8056 | - |
| sidewalk | 89.6064 | - |
| building | 94.8343 | - |
| wall | 71.7874 | - |
| fence | 69.1604 | - |
| pole | 75.7773 | - |
| traffic light | 82.2545 | - |
| traffic sign | 85.4793 | - |
| vegetation | 94.288 | - |
| terrain | 74.9537 | - |
| sky | 96.2557 | - |
| person | 90.245 | 78.9149 |
| rider | 79.817 | 64.8087 |
| car | 96.9965 | 92.5808 |
| truck | 84.3753 | 61.792 |
| bus | 95.7171 | 73.1509 |
| train | 90.5483 | 68.2388 |
| motorcycle | 77.1568 | 63.1791 |
| bicycle | 81.6629 | 73.0369 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.7439 | - |
| nature | 94.0845 | - |
| object | 80.8843 | - |
| sky | 96.2557 | - |
| construction | 94.8895 | - |
| human | 90.2681 | 79.7526 |
| vehicle | 96.6744 | 91.532 |