A smartphone image dataset with lux-meter ground truth for workplace illuminance estimation.

A controlled image dataset of 5814 smartphone photographs is presented for training and benchmarking machine-learning models for indoor illuminance (lux) estimation. All images were acquired using a Samsung Galaxy A16 smartphone mounted on a tripod under fixed camera parameters such as ISO 50, exposure time 1/50 s, white balance 4000 K, 45-degree camera tilt, and 31 cm camera height above the surface. Ground-truth illuminance was recorded using a Lutron LX-101A lux meter at one central point (2300 images) or at five fixed spatial points per image (3514 images). The dataset spans 10 distinct table surfaces and 22 colored paper types photographed across different illuminance levels generated by adjustable LED lamps with correlated color-temperature settings. Surface categories include colored paper (2376 images), white paper (1563 images), real table surfaces (1505 images), and other surfaces (370 images). The dataset enables training and testing of machine learning models for illuminance estimation at defined points on a surface and supports assessment of color classifier performance under varying illuminance conditions. The dataset is publicly deposited in the Zenodo repository (https://doi.org/10.5281/zenodo.20499311).