CVROJan 16, 2025

FLOL: Fast Baselines for Real-World Low-Light Enhancement

arXiv:2501.09718v15 citationsh-index: 6Has Code
Originality Incremental advance
AI Analysis

This addresses efficiency and robustness issues in low-light image enhancement for computational photography applications, though it appears incremental as it builds on existing deep learning methods.

The paper tackles the problem of enhancing low-light images in real-world scenarios, proposing a lightweight neural network that achieves state-of-the-art results on datasets like LOL and LSRW, with processing times under 12ms for 1080p images.

Low-Light Image Enhancement (LLIE) is a key task in computational photography and imaging. The problem of enhancing images captured during night or in dark environments has been well-studied in the image signal processing literature. However, current deep learning-based solutions struggle with efficiency and robustness in real-world scenarios (e.g. scenes with noise, saturated pixels, bad illumination). We propose a lightweight neural network that combines image processing in the frequency and spatial domains. Our method, FLOL+, is one of the fastest models for this task, achieving state-of-the-art results on popular real scenes datasets such as LOL and LSRW. Moreover, we are able to process 1080p images under 12ms. Code and models at https://github.com/cidautai/FLOL

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes