Method Details
Details for method 'GFF-Net'
Method overview
| name | GFF-Net |
| challenge | pixel-level semantic labeling |
| details | We proposed Gated Fully Fusion (GFF) to fuse features from multiple levels through gates in a fully connected way. Specifically, features at each level are enhanced by higher-level features with stronger semantics and lower-level features with more details, and gates are used to control the pass of useful information which significantly reducing noise propagation during fusion. (Joint work: Key Laboratory of Machine Perception, School of EECS @Peking University and DeepMotion AI Research ) |
| publication | GFF: Gated Fully Fusion for Semantic Segmentation Xiangtai Li, Houlong Zhao, Yunhai Tong, Kuiyuan Yang |
| project page / code | |
| used Cityscapes data | fine annotations |
| used external data | ImageNet |
| runtime | n/a |
| subsampling | no |
| submission date | March, 2019 |
| previous submissions |
Average results
| Metric | Value |
|---|---|
| IoU Classes | 82.3181 |
| iIoU Classes | 62.1348 |
| IoU Categories | 92.0247 |
| iIoU Categories | 81.4262 |
Class results
| Class | IoU | iIoU |
|---|---|---|
| road | 98.7381 | - |
| sidewalk | 87.1977 | - |
| building | 93.9065 | - |
| wall | 59.6413 | - |
| fence | 64.3227 | - |
| pole | 71.5199 | - |
| traffic light | 78.3134 | - |
| traffic sign | 82.2344 | - |
| vegetation | 93.9969 | - |
| terrain | 72.5915 | - |
| sky | 95.9379 | - |
| person | 88.2 | 72.7031 |
| rider | 73.9405 | 53.6739 |
| car | 96.4513 | 91.1214 |
| truck | 79.8341 | 45.9651 |
| bus | 92.1587 | 58.0709 |
| train | 84.695 | 56.8141 |
| motorcycle | 71.5266 | 51.5993 |
| bicycle | 78.8371 | 67.1306 |
Category results
| Category | IoU | iIoU |
|---|---|---|
| flat | 98.7633 | - |
| nature | 93.6981 | - |
| object | 77.2065 | - |
| sky | 95.9379 | - |
| construction | 94.2146 | - |
| human | 88.3565 | 73.5506 |
| vehicle | 95.9961 | 89.3018 |