LapLoss: Laplacian Pyramid-based Multiscale loss for Image Translation

TL;DR

LapLoss addresses the challenge of contrast-enhancement in image-to-image translation by preserving fine details across varying illuminations. It introduces a Laplacian pyramid–based translator with level-specific discriminators and a multi-scale loss that combines per-level adversarial and pixel-wise terms, formalized as $L_{total} = \sum_{i=0}^{N} \lambda_i (\mathcal{L}_{GAN}^i + w \mathcal{L}_{MSE}^i)$. The approach yields state-of-the-art SSIM on several SICE benchmarks and notable PSNR gains on mixed-exposure datasets, demonstrating robust texture and structure preservation under challenging lighting. The method balances accuracy and perceptual quality through pyramid-level weighting and lightweight networks, making it effective across underexposed, overexposed, and mixed conditions. These results suggest broad applicability to other I2IT tasks and image restoration domains leveraging Laplacian pyramids and multi-scale adversarial supervision.

Abstract

Contrast enhancement, a key aspect of image-to-image translation (I2IT), improves visual quality by adjusting intensity differences between pixels. However, many existing methods struggle to preserve fine-grained details, often leading to the loss of low-level features. This paper introduces LapLoss, a novel approach designed for I2IT contrast enhancement, based on the Laplacian pyramid-centric networks, forming the core of our proposed methodology. The proposed approach employs a multiple discriminator architecture, each operating at a different resolution to capture high-level features, in addition to maintaining low-level details and textures under mixed lighting conditions. The proposed methodology computes the loss at multiple scales, balancing reconstruction accuracy and perceptual quality to enhance overall image generation. The distinct blend of the loss calculation at each level of the pyramid, combined with the architecture of the Laplacian pyramid enables LapLoss to exceed contemporary contrast enhancement techniques. This framework achieves state-of-the-art results, consistently performing well across different lighting conditions in the SICE dataset.

Figures (3)

• Figure 1: Schematic overview of the LapGSR model kasliwal2024lapgsr employed for multi-level adversarial processing (Fig. \ref{['fig:Generator_Methodl']}). Each branch helps to extract features to finally non-parametrically reconstruct the output image. The instance in the figure is taken from the test set of SICE V1 dataset zheng2024lowlightimagevideoenhancement.
• Figure 2: Schematic overview of the multi-scale GAN paradigm. The affine effect is only applied for visual purposes and is not a preprocessing step. It shows the decomposed pyramid of the predicted image and the ground truth across 3 levels. In the final loss, $\lambda_{1}$, $\lambda_{2}$, and $\lambda_{3}$ are the hyperparameters to weight the level-wise loss explained in Section \ref{['section: loss_method']}.
• Figure 3: Input and output taken for various samples across all datasets. The images in the 1st row are taken from the Overexposure set, 2nd row is taken from Underexposure, 3rd are taken from SICEMix and 4th are taken from SICEGrad. All images are from the test sets.