ILENet: Illumination-Modulated Laplacian-Pyramid Enhancement Network for low-light object detection

With the rapid advancement of deep learning, object detection methods have made significant progress on traditional datasets. However, these methods still face considerable challenges when applied to images captured in low-light conditions. To address this issue, we propose anew method called the Illumination-Modulated Laplacian-Pyramid Enhancement Network (ILENet), specifically for object detection in low-light environments. In ILENet, we design two key components: the Laplacian Enhancement Pyramid (LEP) and the Illumination Modulation Module (IMM). LEP enhances images by optimizing information at different frequencies in low- light images and leveraging local features to improve brightness. IMM further enhances the image by globally querying multi-scale information within the same semantics to generate correction parameters. We integrate the proposed ILENet with a standard YOLO detector, forming anew detection framework. This framework employs a joint training strategy to effectively balance low-light image enhancement (LLIE) and object detection tasks, without requiring paired low-light and normal-light images for pre-training the enhancement network. Quantitative experiments on low-light object detection datasets demonstrate that ILENet outperforms other mainstream LLIE and low-light object detection methods, achieving state-of-the-art (SOTA) performance with an accuracy of 78.5%. Additionally, experimental results on low-light semantic segmentation tasks further validate the effectiveness of ILENet.