Enhanced Low-Light Object Detection Dataset

ExDark RetinexFormer Enhanced Dataset provides annotated low-light images processed with a RetinexFormer enhancement model. It is an adaptation of the original ExDark dataset, formatted for YOLO object detection training. The original ExDark dataset was created by researchers for low-light vision tasks.

Use Cases

• Train YOLO models to detect objects in low-light conditions using the provided bounding box labels.
• Benchmark the performance of object detection algorithms on enhanced versus raw low-light images.
• Fine-tune RetinexFormer or similar enhancement models by using the paired original and enhanced image sets.
• Study the impact of image preprocessing on object detection accuracy in challenging lighting scenarios.

Strengths

• Includes pre-enhanced images using the RetinexFormer model, reducing preprocessing effort.
• Formatted specifically for YOLO training with accompanying dataset.yaml file for configuration.

Limitations

• Unknown number of images and object classes limits assessment of dataset scale and diversity.
• Potential bias from the original ExDark dataset's specific low-light scenarios and object categories.
• Image quality and detection labels depend on the performance of the RetinexFormer enhancement model.

Provenance

Source

Enhanced version derived from the ExDark (Extreme Dark) dataset.

Collection Method

Images processed with a RetinexFormer enhancement model and annotated for YOLO object detection.