Jian Wang, Xiang Long, Guowei Chen, Zewu Wu, Zeyu Chen, Errui Ding et al. "U-HRNet: Delving into Improving Semantic Representation of High Resolution Network for Dense Prediction" arXiv preprint arXiv:2210.07140 (2022).
| Model | Backbone | Resolution | Training Iters | mIoU | mIoU (flip) | mIoU (ms+flip) |
|---|---|---|---|---|---|---|
| FCN | UHRNet_W18_small | 1024x512 | 80000 | 77.66% | 78.26% | 78.47% |
| FCN | UHRNet_W18_small | 1024x512 | 120000 | 78.39% | 79.09% | 79.03% |
| FCN | UHRNet_W48 | 1024x512 | 80000 | 81.28% | 81.76% | 81.48% |
| FCN | UHRNet_W48 | 1024x512 | 120000 | 81.91% | 82.39% | 82.28% |
backbone部分的代码是基于paddleseg修改的
本仓库的架构基于bubbliiiing 的语义分割仓库，使用方法可参考bubbliiiing的视频
数据集格式为VOC Dataset format refers to VOC