Codes for 《 UANet: an Uncertainty-Aware Network for Building Extraction from Remote Sensing Images》

About Paper

We are delighted to inform everyone that our paper has been successfully accepted by IEEE Transactions on Geoscience and Remote Sensing (TGRS 2024). Paper Link

We are delighted to inform you that based on our proposed UANet, we have made improvements and successfully secured the double track championship in the 2024 IEEE GRSS Data Fusion Contest.

The results on the three building datasets can be downloaded via Baidu Disk：Link Code:UANE

We have released the codes of our UANet based on four backbones (VGG, ResNet50, Res2Net-50, and PVT-v2-b2).

The whole training and testing framework of the paper have been released!

Our framwork has been deployed into application: GIF Image

Pretrained Backbones:

We have provided the pretrained backbones(ResNet-50, Res2Net-50, PVT-v2-b2)

You can download via Baidu Disk Link Code:abmg

Training Instructions

To train the UANet model, follow these steps:

Set the CUDA visible devices to specify the GPU for training. For example, to use GPU 0, run the following command:
```
CUDA_VISIBLE_DEVICES=0 python Code/train.py -c config/whubuilding/UANet.py
```

Set the CUDA visible devices to specify the GPU for test. For example, to use GPU 0, run the following command:


CUDA_VISIBLE_DEVICES=0 python Code/test.py -c config/whubuilding/UANet.py -o test_results/whubuilding/UANet/ --rgb

Testing Instructions with Test Time Augmentation (TTA)

To perform testing with Test Time Augmentation (TTA), follow these steps:

Run the following command to perform testing with TTA:


python Code/test.py -c config/whubuilding/UANet.py -o test_results/whubuilding/UANet/ -t lr --rgb

Training Instructions for Multiple Training Sessions

If you want to continue training the model from a checkpoint or perform multiple training sessions, follow this:

Adjust the pretrained_ckpt_path in the config file.

Reference

Our data processing and whole framework are based on the BuildFormer. Here, we sincerely express our gratitude to the authors of that paper.

Citation

We appreciate your attention to our work!


@ARTICLE{10418227,
  author={Li, Jiepan and He, Wei and Cao, Weinan and Zhang, Liangpei and Zhang, Hongyan},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={UANet: An Uncertainty-Aware Network for Building Extraction From Remote Sensing Images}, 
  year={2024},
  volume={62},
  number={},
  pages={1-13},
  keywords={Feature extraction;Uncertainty;Buildings;Data mining;Decoding;Remote sensing;Deep learning;Building extraction;remote sensing (RS);uncertainty-aware},
  doi={10.1109/TGRS.2024.3361211}}

Henryjiepanli / Uncertainty-aware-Network

readme