Method Details
Details for method 'SERNet-Former_v2'
Method overview
| name | SERNet-Former_v2 |
| challenge | pixel-level semantic labeling |
| details | The test results of SERNet-Former could only be acquired partially as this result shows the accuracy of the network on partially tested dataset ranging from Munich's 250th sample to the end. Previously listed as SERNet-Former Partial Result (Munich 250-end). |
| 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 | https://github.com/serdarch/SERNet-Former |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | January, 2024 |
| previous submissions | 1 |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 16.8402 |
| iIoU Classes | 11.825 |
| IoU Categories | 17.1933 |
| iIoU Categories | 13.2974 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 16.0035 | - |
| sidewalk | 11.9402 | - |
| building | 14.1089 | - |
| wall | 6.94112 | - |
| fence | 9.32168 | - |
| pole | 11.3617 | - |
| traffic light | 17.7608 | - |
| traffic sign | 16.1071 | - |
| vegetation | 16.3733 | - |
| terrain | 20.436 | - |
| sky | 14.7695 | - |
| person | 28.0071 | 14.1022 |
| rider | 14.2157 | 9.62846 |
| car | 17.4336 | 11.6353 |
| truck | 19.7195 | 10.8358 |
| bus | 25.3284 | 13.4508 |
| train | 27.5322 | 16.6474 |
| motorcycle | 13.0573 | 7.27801 |
| bicycle | 19.5456 | 11.0216 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 15.7665 | - |
| nature | 16.7766 | - |
| object | 13.9349 | - |
| sky | 14.7695 | - |
| construction | 13.9745 | - |
| human | 26.8043 | 14.3483 |
| vehicle | 18.3269 | 12.2465 |