Low-light image dataset preserves alignment across night levels
3L-D: A Low-Light Level Dataset from a Sensor-Agnostic, Alignment-Preserving Acquisition Framework
Computer Vision and Pattern RecognitionDatabases
Summary
Taking clear photos or videos in very dark conditions is hard because existing datasets don’t capture true nighttime details or noise well. The authors created a new dataset called 3L-D using a device that can simulate different night brightness levels while keeping images perfectly aligned. This dataset covers realistic night scenes from bright moonlight to dark overcast nights and includes near-infrared light. They tested several AI models on this dataset, showing how quality and speed trade off when enhancing very dark images.
What this means in practice
- •For computer vision developers: Train and evaluate low-light image enhancement models using a dataset with aligned images across realistic night brightness levels.
- •For surveillance camera engineers: Design and test night-vision systems benefitting from sensor-agnostic data simulating diverse nighttime environments with realistic noise.
Authors
Léo Rueff, Maurgène Ghezal, Valentin Gorse, Astrid Dufaure, Esteban Monsalvo, François Atteia, Ludovic Escoubas
Abstract
Supervised low-light image and video enhancement has achieved remarkable progress, yet current models remain fundamentally constrained by the limitations of existing datasets. Prevailing methods typically rely on uncontrolled ambient captures, camera parameter manipulation (e.g., exposure or ISO adjustments), or neutral density filters, failing to replicate authentic night-sky near-infrared spectral content and genuine low-photon noise distributions. Furthermore, capturing dynamic paired sequences without introducing warping artifacts or sensor-specific biases remains an open challenge. In this paper, we introduce 3L-Dataset, a physically calibrated low-light image and video dataset generated using a modular, sensor-agnostic acquisition framework. Our test bench faithfully reproduces the standardized Midavaine night level scale from Level 1 (full moon, approximately 1000 mlx) down to Level 5 (overcast moonless night, approximately 0.1 mlx), spanning the visible to near-infrared spectrum across diverse realistic environments, including countryside, forest, urban, and desert scale models. Using high-precision programmable mechanical actuation, our acquisition framework guarantees sub-pixel spatial alignment across synchronized sextuplets, each composed of one reference image and five corresponding images acquired at the five standardized night levels, without post-processing image warping. We evaluate multiple representative deep learning baselines across varying night levels, revealing critical trade-offs between reconstruction fidelity, perceptual quality, and inference latency under severe photon starvation. The modular acquisition framework can be adapted to different cameras, optical systems, sensor technologies, and resolutions, enabling the dataset to be continuously expanded.