Spatial-frequency Dual-Domain Feature Fusion Network for Low-Light Remote Sensing Image Enhancement

This repository is an official PyTorch implementation of the paper Spatial-frequency Dual-Domain Feature Fusion Network for Low-Light Remote Sensing Image Enhancement.

Fig. 1. Comparison between the latest state-of-the-art methods and our approach.

☑️ TODO

[x] Build the repo
[x] arXiv version
[x] Dataset
[x] Release code
[x] Pretrained weights&log_files
[x] Add Download Link for Visual Results on Common Benckmarks

🔍Dataset

We proposed two datasets iSAID-dark and darkrs. Please click ISAID and darkrs for detailed preparation description.

Fig. 2. Samples from the proposed iSAID-dark(Up) and darkrs(Down) dataset.

Training & Testing

python test.py

🔍Visual Results

Fig. 3. The visualization results on the iSAID-dark dataset. We present the histogram of color distribution for the images. The histograms placed in Input/GT represent the color distribution of the GT. It can be observed that our method’s histogram is closer to the GT histogram.

Fig. 4. The visualization results on the DICM dataset (top) and the NPE dataset (bottom).

✈️ Citation

@article{yao2024spatial,
  title={Spatial-frequency dual-domain feature fusion network for low-light remote sensing image enhancement},
  author={Yao, Zishu and Fan, Guodong and Fan, Jinfu and Gan, Min and Chen, CL Philip},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2024},
  publisher={IEEE}
}

iijjlk / DFFN