FAEWNet

This is a PyTorch implementation of the paper SAM-Based Building Change Detection with Distribution-Aware Fourier Adaptation and Edge-Constrained Warping

simple usage

pip install -U openmim
mim install mmengine
mim install "mmcv>=2.0.0"
mim install "mmpretrain>=1.0.0rc7"
pip install "mmsegmentation>=1.0.0rc6"
pip install "mmdet>=3.0.0"
cd  FAEWNet-main
pip install -v -e .
pip install peft

Dataset preparation

      Data
      ├── LEVIR-CD
      │   ├── train
      │   │   ├── A
      │   │   ├── B
      │   │   ├── label
      │   ├── val
      │   │   ├── A
      │   │   ├── B
      │   │   ├── label
      │   ├── test
      │   │   ├── A
      │   │   ├── B
      │   │   ├── label
      │   ├── list
      │   │   ├── train.txt
      │   │   ├── val.txt
      │   │   ├── test.txt

Train

python tools/train.py configs/faewnet/faewnet_vit-sam-l_512x512_fp16-719e_levircd.py --work-dir ./faewnet_levircd

python tools/train.py configs/faewnet/faewnet_vit-sam-l_512x512_fp16-80k_s2looking.py --work-dir ./faewnet_s2looking

python tools/train.py configs/faewnet/faewnet_vit-sam-l_256x256_fp16-100e_whucd.py --work-dir ./faewnet_whucd

Infer

python tools/test.py configs/faewnet/faewnet_vit-sam-l_512x512_fp16-719e_levircd.py faewnet_levircd/best_mIoU_epoch_595.pth 

python tools/test.py configs/faewnet/faewnet_vit-sam-l_512x512_fp16-80k_s2looking.py faewnet_s2looking/best_mIoU_iter_72000.pth

python tools/test.py configs/faewnet/faewnet_vit-sam-l_256x256_fp16-100e_whucd.py faewnet_whucd/best_mIoU_epoch_80.pth 

Results

Visual comparison of building change results on the WHU-CD dataset.

(a) T1 image, (b) T2 image, (c) Change label, (d) ChangerAD (ResNet18), (e) Changer (ResNet18), (f) IDA-SiamNet (ResNet18), (g) BiT, (h) ChangeFormer, (i) ChangerAD (MiT-b1), (j) Changer (MiT-b1), (k) IDA-SiamNet (MiT-b1), (l) BAN, (m) TTP and (n) FAEWNet. True positive is indicated in white, true negative in black, false positive in red, and false negative in blue.

Pretrained Models

For readers to easily evaluate the accuracy, we provide the trained weights of the FAEWNet

Baidu (pswd: 8jih)

Citation

If you find this project useful in your research, please consider cite:

@ARTICLE{11227013,
  author={Li, Yun-Cheng and Lei, Sen and Zhao, Yi-Tao and Li, Heng-Chao and Li, Jun and Plaza, Antonio},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={SAM-Based Building Change Detection With Distribution-Aware Fourier Adaptation and Edge-Constrained Warping}, 
  year={2025},
  volume={63},
  number={},
  pages={1-14},
  doi={10.1109/TGRS.2025.3629110}}