UNICE: Training A Universal Image Contrast Enhancer
This addresses the need for a generalized model for various image contrast enhancement tasks, such as under-/over-exposure correction and low-light enhancement, which is incremental as it builds on existing multi-exposure fusion techniques.
The authors tackled the problem of poor generalization in image contrast enhancement methods across different tasks by proposing UNICE, a universal enhancer that generates multi-exposure sequences from a single sRGB image and fuses them, achieving significantly stronger generalization performance and outperforming manually created ground-truths in multiple no-reference image quality metrics.
Existing image contrast enhancement methods are typically designed for specific tasks such as under-/over-exposure correction, low-light and backlit image enhancement, etc. The learned models, however, exhibit poor generalization performance across different tasks, even across different datasets of a specific task. It is important to explore whether we can learn a universal and generalized model for various contrast enhancement tasks. In this work, we observe that the common key factor of these tasks lies in the need of exposure and contrast adjustment, which can be well-addressed if high-dynamic range (HDR) inputs are available. We hence collect 46,928 HDR raw images from public sources, and render 328,496 sRGB images to build multi-exposure sequences (MES) and the corresponding pseudo sRGB ground-truths via multi-exposure fusion. Consequently, we train a network to generate an MES from a single sRGB image, followed by training another network to fuse the generated MES into an enhanced image. Our proposed method, namely UNiversal Image Contrast Enhancer (UNICE), is free of costly human labeling. However, it demonstrates significantly stronger generalization performance than existing image contrast enhancement methods across and within different tasks, even outperforming manually created ground-truths in multiple no-reference image quality metrics. The dataset, code and model are available at https://github.com/BeyondHeaven/UNICE.