Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging Robustness and Efficiency: Real-Time Low-Light Enhancement via Attention U-Net GAN

Yash Thesia, Meera Suthar

TL;DR

This work addresses the real-time low-light enhancement problem by bridging the gap between fast CNN-based methods and high-fidelity diffusion models. It introduces an Attention U-Net GAN that processes raw Bayer data in a single forward pass, using Attention Gates and adversarial training to recover high-frequency textures without iterative sampling. On the SID dataset, it achieves LPIPS of 0.112 with latency around 0.062s, outperforming efficient baselines and offering substantial speedups over diffusion models. The approach provides a practical, edge-deployable solution with perceptual quality close to generative methods, enabling real-time applications in mobile photography and video processing.

Abstract

Recent advancements in Low-Light Image Enhancement (LLIE) have focused heavily on Diffusion Probabilistic Models, which achieve high perceptual quality but suffer from significant computational latency (often exceeding 2-4 seconds per image). Conversely, traditional CNN-based baselines offer real-time inference but struggle with "over-smoothing," failing to recover fine structural details in extreme low-light conditions. This creates a practical gap in the literature: the lack of a model that provides generative-level texture recovery at edge-deployable speeds. In this paper, we address this trade-off by proposing a hybrid Attention U-Net GAN. We demonstrate that the heavy iterative sampling of diffusion models is not strictly necessary for texture recovery. Instead, by integrating Attention Gates into a lightweight U-Net backbone and training within a conditional adversarial framework, we can approximate the high-frequency fidelity of generative models in a single forward pass. Extensive experiments on the SID dataset show that our method achieves a best-in-class LPIPS score of 0.112 among efficient models, significantly outperforming efficient baselines (SID, EnlightenGAN) while maintaining an inference latency of 0.06s. This represents a 40x speedup over latent diffusion models, making our approach suitable for near real-time applications.

Bridging Robustness and Efficiency: Real-Time Low-Light Enhancement via Attention U-Net GAN

TL;DR

This work addresses the real-time low-light enhancement problem by bridging the gap between fast CNN-based methods and high-fidelity diffusion models. It introduces an Attention U-Net GAN that processes raw Bayer data in a single forward pass, using Attention Gates and adversarial training to recover high-frequency textures without iterative sampling. On the SID dataset, it achieves LPIPS of 0.112 with latency around 0.062s, outperforming efficient baselines and offering substantial speedups over diffusion models. The approach provides a practical, edge-deployable solution with perceptual quality close to generative methods, enabling real-time applications in mobile photography and video processing.

Abstract

Recent advancements in Low-Light Image Enhancement (LLIE) have focused heavily on Diffusion Probabilistic Models, which achieve high perceptual quality but suffer from significant computational latency (often exceeding 2-4 seconds per image). Conversely, traditional CNN-based baselines offer real-time inference but struggle with "over-smoothing," failing to recover fine structural details in extreme low-light conditions. This creates a practical gap in the literature: the lack of a model that provides generative-level texture recovery at edge-deployable speeds. In this paper, we address this trade-off by proposing a hybrid Attention U-Net GAN. We demonstrate that the heavy iterative sampling of diffusion models is not strictly necessary for texture recovery. Instead, by integrating Attention Gates into a lightweight U-Net backbone and training within a conditional adversarial framework, we can approximate the high-frequency fidelity of generative models in a single forward pass. Extensive experiments on the SID dataset show that our method achieves a best-in-class LPIPS score of 0.112 among efficient models, significantly outperforming efficient baselines (SID, EnlightenGAN) while maintaining an inference latency of 0.06s. This represents a 40x speedup over latent diffusion models, making our approach suitable for near real-time applications.
Paper Structure (12 sections, 5 equations, 2 figures, 3 tables)

This paper contains 12 sections, 5 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset
  4. Proposed Method
  5. Network Architecture
  6. Objective Function
  7. Experiments
  8. Experimental Settings
  9. Comparative Evaluation
  10. Ablation Study
  11. Qualitative Results
  12. Conclusion and Future Work

Figures (2)

  • Figure 1: Proposed Attention U-Net GAN Architecture. The generator processes raw Bayer data packed into 4 channels, utilizing Attention Gates to suppress noise in dark regions. The discriminator operates on patches to enforce texture realism.
  • Figure 2: Qualitative Analysis. Visual comparison of our Attention U-Net GAN against the SID baseline Chen2018LearningTS. Our method (c, g, k) recovers better contrast and color fidelity compared to the baseline, with attention maps (d, h, l) highlighting the active regions of enhancement.