The paper introduces a novel method called Zero-Reference Deep Curve Estimation (Zero-DCE) for enhancing low-light images. Zero-DCE formulates the light enhancement task as an image-specific curve estimation problem using a lightweight deep network, DCE-Net. The method trains without requiring any paired or unpaired reference images through a set of non-reference loss functions, including spatial consistency, exposure control, color constancy, and illumination smoothness losses. These losses implicitly measure the quality of the enhanced images and guide the network's learning. The proposed method is efficient, capable of real-time processing, and generalizes well to diverse lighting conditions. Extensive experiments on various benchmarks demonstrate the superior performance of Zero-DCE over state-of-the-art methods in both qualitative and quantitative metrics. Additionally, the method shows potential benefits in improving face detection in low-light conditions.The paper introduces a novel method called Zero-Reference Deep Curve Estimation (Zero-DCE) for enhancing low-light images. Zero-DCE formulates the light enhancement task as an image-specific curve estimation problem using a lightweight deep network, DCE-Net. The method trains without requiring any paired or unpaired reference images through a set of non-reference loss functions, including spatial consistency, exposure control, color constancy, and illumination smoothness losses. These losses implicitly measure the quality of the enhanced images and guide the network's learning. The proposed method is efficient, capable of real-time processing, and generalizes well to diverse lighting conditions. Extensive experiments on various benchmarks demonstrate the superior performance of Zero-DCE over state-of-the-art methods in both qualitative and quantitative metrics. Additionally, the method shows potential benefits in improving face detection in low-light conditions.