Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

IAML: Illumination-Aware Mirror Loss for Progressive Learning in Low-Light Image Enhancement Auto-encoders

Farida Mohsen, Tala Zaim, Ali Al-Zawqari, Ali Safa, Samir Belhaouari

Abstract

This letter presents a novel training approach and loss function for learning low-light image enhancement auto-encoders. Our approach revolves around the use of a teacher-student auto-encoder setup coupled to a progressive learning approach where multi-scale information from clean image decoder feature maps is distilled into each layer of the student decoder in a mirrored fashion using a newly-proposed loss function termed Illumination-Aware Mirror Loss (IAML). IAML helps aligning the feature maps within the student decoder network with clean feature maps originating from the teacher side while taking into account the effect of lighting variations within the input images. Extensive benchmarking of our proposed approach on three popular low-light image enhancement datasets demonstrate that our model achieves state-of-the-art performance in terms of average SSIM, PSNR and LPIPS reconstruction accuracy metrics. Finally, ablation studies are performed to clearly demonstrate the effect of IAML on the image reconstruction accuracy.

IAML: Illumination-Aware Mirror Loss for Progressive Learning in Low-Light Image Enhancement Auto-encoders

Abstract

This letter presents a novel training approach and loss function for learning low-light image enhancement auto-encoders. Our approach revolves around the use of a teacher-student auto-encoder setup coupled to a progressive learning approach where multi-scale information from clean image decoder feature maps is distilled into each layer of the student decoder in a mirrored fashion using a newly-proposed loss function termed Illumination-Aware Mirror Loss (IAML). IAML helps aligning the feature maps within the student decoder network with clean feature maps originating from the teacher side while taking into account the effect of lighting variations within the input images. Extensive benchmarking of our proposed approach on three popular low-light image enhancement datasets demonstrate that our model achieves state-of-the-art performance in terms of average SSIM, PSNR and LPIPS reconstruction accuracy metrics. Finally, ablation studies are performed to clearly demonstrate the effect of IAML on the image reconstruction accuracy.
Paper Structure (11 sections, 10 equations, 1 figure, 4 tables)

This paper contains 11 sections, 10 equations, 1 figure, 4 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Teacher-student auto-encoder setup
  4. Illumination-aware mirror loss
  5. Global loss function
  6. Experimental setup
  7. Model architecture and training procedure
  8. Benchmark Datasets
  9. Benchmark results
  10. Ablation studies
  11. Conclusion

Figures (1)

  • Figure 1: Proposed teacher-student auto-encoder setup for low-light image enhancement. The student network $\mathcal{S}$ receives as input the degraded low-light images $I_{\text{low-light}}$ and seeks to recover the clean images $\hat{I}_{\text{clean}}$. The teacher network $\mathcal{T}$ shares the same encoder as $\mathcal{S}$ (updated as $\mathcal{T}_E\xleftarrow[]{} \mathcal{S}_E$ during the training iterations) and updates its decoder $\mathcal{T}_D$ via an exponential moving average using the weights of the student decoder $\mathcal{S}_D$ (no gradient-based learning in $\mathcal{T}$). The teacher takes as input the clean target images $I_{\text{clean}}$, leading to clean multi-scale feature maps at each layer of $\mathcal{T}_D$. These clean feature maps are then used to guide the learning of the layers within $\mathcal{S}_D$ in a mirrored fashion using the proposed Illumination-Aware Mirror Loss (IAML).