Striving for Faster and Better: A One-Layer Architecture with Auto Re-parameterization for Low-Light Image Enhancement
Nan An, Long Ma, Guangchao Han, Xin Fan, RIsheng Liu
TL;DR
This paper addresses the challenge of achieving high-quality low-light image enhancement with real-time efficiency by proposing AR-LLIE, a one-layer CNN augmented with auto re-parameterization and guided by tiered neural architecture search. By learning a rich multi-branch representation during training and collapsing it into a single 3×3 convolution at inference, the method delivers state-of-the-art visual quality while maintaining exceptional speed on CPU, GPU, DSP, and NPU. The authors introduce a formal loss framework combining fidelity and spatial smoothing, and validate the approach across multiple datasets (MIT, LOL, DARKFACE, BAID) with extensive ablations and a downstream detection study showing practical benefits. The work demonstrates that task-aware re-parameterization, when coupled with structured architecture search, can push the limits of efficiency without sacrificing perceptual quality, and provides evidence of broad applicability in real-time imaging systems.
Abstract
Deep learning-based low-light image enhancers have made significant progress in recent years, with a trend towards achieving satisfactory visual quality while gradually reducing the number of parameters and improving computational efficiency. In this work, we aim to delving into the limits of image enhancers both from visual quality and computational efficiency, while striving for both better performance and faster processing. To be concrete, by rethinking the task demands, we build an explicit connection, i.e., visual quality and computational efficiency are corresponding to model learning and structure design, respectively. Around this connection, we enlarge parameter space by introducing the re-parameterization for ample model learning of a pre-defined minimalist network (e.g., just one layer), to avoid falling into a local solution. To strengthen the structural representation, we define a hierarchical search scheme for discovering a task-oriented re-parameterized structure, which also provides powerful support for efficiency. Ultimately, this achieves efficient low-light image enhancement using only a single convolutional layer, while maintaining excellent visual quality. Experimental results show our sensible superiority both in quality and efficiency against recently-proposed methods. Especially, our running time on various platforms (e.g., CPU, GPU, NPU, DSP) consistently moves beyond the existing fastest scheme. The source code will be released at https://github.com/vis-opt-group/AR-LLIE.