A New Dual-Scale Hyperspectral Dataset.

This work proposes a new high-quality and high-resolution hyperspectral dataset captured using a Specim IQ hyperspectral line scanner. It includes 23 images with and without colour calibration target (46 images), including wide framing and closeups of different salient objects and materials in each scene. The dataset is intended for training and testing illumination estimation algorithms, as well as techniques for super-resolution and spectral reconstruction. For each image, RAW sensor response, spectral reference values, estimated light source, calculated reflectance data, RGB thumbnails, and metadata are provided. The dataset also includes ground-truth data obtained from the spectral reflectances of each reference patch.