🤖 AI Summary
Low-light imaging severely degrades downstream vision tasks due to significant illumination deficiency. To address this, we propose the Frequency-domain Radial Basis Network (FRBNet), the first framework to model low-light enhancement explicitly in the frequency domain and formulate it as an end-to-end trainable pipeline grounded in an extended Lambertian reflectance model. Our key contributions are: (1) a theoretical frequency-domain channel-ratio principle that characterizes illumination-invariant feature responses; and (2) a learnable, structured frequency-domain filtering module enabling plug-and-play feature enhancement without architectural or loss-function modifications. Extensive experiments on low-light object detection and nighttime semantic segmentation demonstrate consistent improvements—+2.2 mAP and +2.9 mIoU—validating FRBNet’s effectiveness, generalizability, and practical plug-and-play utility across diverse vision tasks.
📝 Abstract
Low-light vision remains a fundamental challenge in computer vision due to severe illumination degradation, which significantly affects the performance of downstream tasks such as detection and segmentation. While recent state-of-the-art methods have improved performance through invariant feature learning modules, they still fall short due to incomplete modeling of low-light conditions. Therefore, we revisit low-light image formation and extend the classical Lambertian model to better characterize low-light conditions. By shifting our analysis to the frequency domain, we theoretically prove that the frequency-domain channel ratio can be leveraged to extract illumination-invariant features via a structured filtering process. We then propose a novel and end-to-end trainable module named extbf{F}requency-domain extbf{R}adial extbf{B}asis extbf{Net}work ( extbf{FRBNet}), which integrates the frequency-domain channel ratio operation with a learnable frequency domain filter for the overall illumination-invariant feature enhancement. As a plug-and-play module, FRBNet can be integrated into existing networks for low-light downstream tasks without modifying loss functions. Extensive experiments across various downstream tasks demonstrate that FRBNet achieves superior performance, including +2.2 mAP for dark object detection and +2.9 mIoU for nighttime segmentation. Code is available at: https://github.com/Sing-Forevet/FRBNet.