The Exclusively Dark (ExDARK) dataset is one of the largest curated collections of real-world low-light images designed to support research in computer vision tasks under challenging lighting conditions. It contains 7,363 images captured across ten different low-light scenarios, ranging from extremely dark environments to twilight. Each image is annotated with both image-level labels and object-level bounding boxes for 12 object categories, making it suitable for detection and classification tasks. The dataset was created to address the lack of large-scale low-light datasets available for research in object detection, recognition, and enhancement. It has been widely used in studies of low-light image enhancement, deep learning approaches, and domain adaptation for vision models. Researchers can also explore its associated source code for low-light image enhancement tasks, making it an essential resource for advancing work in night-time and low-light visual recognition.
Features
- Contains 7,363 low-light images across 10 distinct lighting conditions
- Provides annotations at both the image class and object bounding box levels
- Includes 12 object categories similar to PASCAL VOC standard classes
- Supports tasks in object detection, recognition, and image enhancement
- Designed for research in deep learning, domain adaptation, and low-light vision
- Open dataset under BSD-3-Clause license with published academic references
Project Samples
