Introduction to the Exclusively Dark (ExDark) Image Dataset
The Exclusively Dark (ExDark) Image Dataset is a groundbreaking resource designed to advance research in object detection and image enhancement within low-light environments. This dataset is particularly intended for use in computer vision tasks and has been recognized in the field for its unique contributions.
Dataset Overview
Originally released on May 29, 2018, the ExDark dataset was developed to fill a gap in resources available for studying images captured in low-light settings. The dataset was created by Yuen Peng Loh and Chee Seng Chan and has been published in the Computer Vision and Image Understanding journal. It consists of 7,363 images taken under various low-light conditions that range from very dim environments to twilight.
These images are categorized into 12 object classes, closely resembling the categories found in the popular PASCAL VOC dataset. Each image in the ExDark collection comes with annotations that detail both the class of the image and the bounding boxes for local objects within these images.
Purpose and Applications
The primary aim of the ExDark dataset is to support and expand research in domains that require understanding and processing low-light images. This is particularly challenging, yet essential, as many real-world applications, such as nighttime surveillance and automotive safety systems, frequently encounter such conditions. Researchers and practitioners can leverage this dataset to develop new algorithms for object detection and image enhancement, potentially leading to significant advancements in these areas.
Source Code
To assist researchers working with the ExDark dataset, the team has also released source code for low-light image enhancement. This code is available in the SPIC folder within their GitHub repository, providing a practical starting point for those looking to improve or develop applications focused on enhancing images captured under poor lighting conditions.
Citation and Acknowledgments
Researchers using the ExDark dataset are encouraged to cite the work as follows:
@article{Exdark,
title = {Getting to Know Low-light Images with The Exclusively Dark Dataset},
author = {Loh, Yuen Peng and Chan, Chee Seng},
journal = {Computer Vision and Image Understanding},
volume = {178},
pages = {30-42},
year = {2019},
doi = {https://doi.org/10.1016/j.cviu.2018.10.010}
}
Feedback and suggestions for the dataset are highly welcomed by the authors, who are keen to hear both positive and negative opinions. To provide feedback, individuals can contact the authors via email.
Licensing and Usage
The ExDark dataset is an open-source project made available under the BSD-3 license. However, for any commercial use, users are requested to contact Dr. Chee Seng Chan for permissions. The dataset and its associated resources have been developed and maintained by the Center of Image and Signal Processing at the Faculty of Computer Science and Information Technology, Universiti Malaya.
For researchers and developers working in the domain of low-light image processing, the ExDark dataset is an invaluable tool providing both a challenge and an opportunity to drive innovation and improve techniques in this essential area of computer vision.