Method Details
Details for method 'SERNet-Former-woAt'
Method overview
| name | SERNet-Former-woAt |
| challenge | pixel-level semantic labeling |
| details | SERNet-Former Ablation Study: SERNET-Former is developed by using ResNet-50 as the baseline. This result shows the efficiency of the baseline itself trained for 60 epochs without using the additional methods. |
| publication | SERNet-Former: Semantic Segmentation by Efficient Residual Network with Attention-Boosting Gates and Attention-Fusion Networks Serdar Erisen https://doi.org/10.48550/arXiv.2401.15741 |
| project page / code | |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | January, 2024 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 69.6534 |
| iIoU Classes | 45.1343 |
| IoU Categories | 86.7558 |
| iIoU Categories | 72.2926 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 97.0643 | - |
| sidewalk | 76.371 | - |
| building | 89.482 | - |
| wall | 46.5399 | - |
| fence | 47.9099 | - |
| pole | 55.0053 | - |
| traffic light | 62.9756 | - |
| traffic sign | 69.007 | - |
| vegetation | 91.2711 | - |
| terrain | 67.1256 | - |
| sky | 94.0108 | - |
| person | 79.7119 | 59.0404 |
| rider | 61.2579 | 40.5148 |
| car | 93.0889 | 85.3141 |
| truck | 50.8364 | 22.297 |
| bus | 62.4244 | 33.6118 |
| train | 54.2383 | 32.2961 |
| motorcycle | 57.7583 | 32.7868 |
| bicycle | 67.335 | 55.2134 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.0706 | - |
| nature | 90.809 | - |
| object | 62.3802 | - |
| sky | 94.0108 | - |
| construction | 89.8719 | - |
| human | 80.3342 | 60.9581 |
| vehicle | 91.8139 | 83.627 |