AE-YOLO: Asymptotic Enhancement for Low-Light Object Detection

Object detection has achieved significant advancements due to the continuous development of deep learning technology; however, object detection in low-light conditions remains a challenge. To address this issue, we propose an asymptotic enhancement network (AENet) and integrate it with YOLOv3 to develop a novel object detection framework called AE-YOLO, specially designed for low-light. To maximize the benefits of the enhancement network for downstream detection tasks, AENet employs pixel-level enhancement and feature-level enhancement to adaptively enhance an image to improve detection performance. Specifically, the pixel-level enhancement network first divides an image into <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$8 \times 8$</tex> patches, and applies the Yeo-Johnson transform on each patch for dynamic enhancement. The feature-level enhancement network utilizes a prompt block to recover details lost in dark, followed by a transformer layer for further feature enhancement. We evaluated the performance of AE-YOLO using the ExDark dataset, which is specifically designed for low-light object detection. The experimental results demonstrate the effectiveness of our proposed AE-YOLO framework in enhancing image quality and improving object detection accuracy under low-light conditions.