Synthetic Image Restoration Dataset for Hazy, Low-Light, and Rainy Images

Description

This dataset contains 2,154 original Ground Truth images and corresponding synthetically degraded images generated for three common image degradation conditions: Low-light, Hazy, and Rainy. For each of the 2,154 Ground Truth images, three corresponding degraded versions are provided: 2,154 Low-light images, 2,154 Hazy images, 2,154 Rainy images. Thus, the dataset contains 8,616 images in total, consisting of 2,154 clean Ground Truth images and 6,462 synthetically degraded images. Each degraded image is paired with its corresponding Ground Truth image from the same original scene, making the dataset suitable for supervised image restoration and enhancement research. The dataset can be used to develop and evaluate unified restoration models for multiple degradation types, including low-light enhancement, image dehazing, and image deraining. The dataset was prepared to support research in deep learning, computer vision, and image restoration, and can be used for model training, validation, testing, comparative evaluation, and further research on multi-degradation image restoration.

Steps to reproduce

DATA PRODUCTION PROCESS Primary Multi-Degradation Image Restoration Dataset Dataset Overview This dataset contains original Ground Truth images and corresponding synthetically degraded images representing three common degradation conditions: Low-light Hazy Rainy Each Ground Truth image has a corresponding degraded version for each condition. This paired structure makes the dataset suitable for supervised image restoration and enhancement research. Data Production Steps 1. Collection of Clean Images Original clean images were collected and used as the Ground Truth data for the dataset. 2. Ground Truth Organization The clean images were organized in a dedicated Ground Truth folder. Their original filenames were retained to maintain correspondence with the degraded images. 3. Synthetic Low-Light Generation The clean images were modified by reducing illumination or brightness to simulate low-light conditions. 4. Synthetic Haze Generation The clean images were processed using a synthetic haze-generation method to simulate atmospheric haze or fog. 5. Synthetic Rain Generation Synthetic rain patterns or streaks were added to the clean images to simulate rainy conditions. 6. Image Pairing Each degraded image was matched with its corresponding Ground Truth image using the same filename. For example: Ground Truth/image (001).jpg Low-light/image (001).jpg Hazy/image (001).jpg Rainy/image (001).jpg These files represent the same underlying scene under different conditions. 7. Dataset Verification The generated dataset was checked to ensure that: Ground Truth and degraded images are correctly aligned. Each degraded image has a corresponding Ground Truth image. Filenames remain consistent across categories. No expected image is missing. 8. Final Organization After verification, the images were organized into separate folders for Ground Truth, Low-light, Hazy, and Rainy data. Dataset Structure Dataset/ ├── Ground Truth/ ├── Low-light/ ├── Hazy/ └── Rainy/ Each folder contains images representing the corresponding category. Data Relationship Original Clean Image │ ├── Ground Truth ├── Low-light ├── Hazy └── Rainy The dataset provides paired clean and degraded images for supervised image restoration. Intended Research Use The dataset can support research in: Computer Vision Machine Learning Deep Learning Image Processing Image Restoration Image Enhancement Low-Light Enhancement Image Dehazing Image Deraining Multi-Degradation Image Restoration Important Note The Low-light, Hazy, and Rainy images are synthetically generated from clean images. Therefore, they simulate common degradation conditions but may not represent the full complexity of naturally degraded real-world images. The dataset is intended primarily for research, training, validation, and evaluation of image restoration and enhancement models.

Institutions

• Daffodil International University

  Dhaka Division, Dhaka

Categories

Licence