Published in ACM Multimedia 2024, Melbourne, Australia

HazeSpace2M: A Dataset for Haze Aware Single Image Dehazing [Paper]

Md Tanvir Islam ¹, Nasir Rahim ¹, Saeed Anwar ², Muhammad Saqib ³, Sambit Bakshi ⁴, Khan Muhammad ^{1, *}

| 1. Sungkyunkwan University, South Korea | 2. ANU, Australia | 3. CSIRO, Australia | 4. NIT Rourkela, India || *Corresponding Author |

---

IMPORTANT UPDATES

• 2025/02/24 | Checked: No indexing issues were found in the Street subset.
• 2025/02/24 | Fixed the indexing issues of the Outdoor subset.
• 2025/02/22 | Fixed the indexing issues of the Farmland subset.
• 2025/02/21 | Identified an indexing issue between GT and Haze images. We are working on it to fix the issues.

---

HazeSpace2M Dataset

Haze Aware Dehazing

Dependencies

pip install -r requirements.txt

Dataset Download

The subsets of the HazeSpace2M dataset are available for download.

1. Outdoor (269GB) | 2. Street (295GB) | 3. Farmland (90GB) | 4. Satellite (153GB)

Pretrained Weights

All the pre-trained weights of the classifiers and the dehazers are available to download:
Google Drive: | Classifier | Specialized Dehazers |

Testing

python inference.py --gt_folder <path_to_gt> --hazy_folder <path_to_hazy> --output_dir <output_dir> --classifier <path_to_classifier> --cloudSD <path_to_cloudSD> --ehSD <path_to_ehSD> --fogSD <path_to_fogSD>

Note: Each variable is explained in the inference.py file.

Use Own Classifiers

To use your custom classifier, please follow the following steps:

1. Write the code for your classifier architecture in the classifier.py file in the models folder.
2. Now define the object of your classifier in the classification_inference method inside the conditionalDehazing.py file under the models folder.
3. Finally, define the weights of your classifier inside the inference.py file

To use your custom specialized dehazers, please follow the following steps:

1. Write the code for your classifier architecture in the dehazer.py file in the models folder.
2. Now define the object of your dehazer in the load_model method inside the helper.py file under the utils folder.
3. Finally, define the weights of your classifier inside the inference.py file

HazeSpace2M Folder Structure

Cite this Paper

If you find our work useful in your research, please consider citing our paper and star ✨✨ this repository. Thank you!

@inproceedings{hazespace2m,
  title={HazeSpace2M: A Dataset for Haze Aware Single Image Dehazing},
  author={Islam, Md Tanvir and Rahim, Nasir and Anwar, Saeed and Saqib Muhammad and Bakshi, Sambit and Muhammad, Khan},
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},
  year={2024},
  doi = {10.1145/3664647.3681382}
}