SMALL: Smartphone Multi-Device Low-Light Dataset
Description
SMALL (Smartphone Multi-Device Low-Light Dataset) is a paired smartphone image dataset for supervised low-light image enhancement. The dataset contains 504 low-light/brighter-reference pairs (1,008 images) from 371 indoor scenes captured using five smartphone models. The images are high-resolution rendered-sRGB files rather than sensor RAW data. The release provides paired images, scene identifiers, device and image geometry metadata, train/validation/test split annotations, and dataset-relative underexposure severity labels derived from scene-level low-image luminance proxies. The recommended scene-level split contains 403 training pairs, 51 validation pairs, and 50 test pairs, with no scene shared across subsets. The dataset is intended for supervised low-light image enhancement, severity-aware evaluation, cross-device analysis, robustness studies, and qualitative failure analysis. The release also includes metadata, split manifests, reproducibility scripts, integrity checks, and SHA-256 checksums. Version: 1.0.0 License: Creative Commons Attribution 4.0 International (CC BY 4.0)
Files
Institutions
- FPT UniversityHanoi, Hanoi