Benchmarking Single Image Dehzing Methods Under Realistic Sunlight Haze

Sun-Haze dataset is a benchmark dataset for single image dehazing that include 112 hazy images with realistic sunlight haze. Since the ground truth of a hazy image could be a range of clean images, this dataset include 6 ground truth images (one original image plus 5 images retouched by 5 experts) per hazy image to provide the opportunity to test image dehazing methods in a more practical way. This dataset is built on top of MIT-Adobe FiveK dataset. Paper is here.

Sun-Haze Dataset

Sun-Haze dataset is available here.

Analysis Results

Quantitative Results

alt text

Qualitative Results

alt text

Publication

If you find this work useful for you, please cite:

@inproceedings{anvari2020evaluating,
  title={Evaluating Single Image Dehazing Methods Under Realistic Sunlight Haze},
  author={Anvari, Zahra and Athitsos, Vassilis},
  booktitle={International Symposium on Visual Computing},
  pages={436--447},
  year={2020},
  organization={Springer}
}