HyperTex-Splits: Standardized Train-Test Partitions for Fibre Composition Estimation

HyperTex-Splits is a companion dataset to the HyperTex hyperspectral textile dataset. It provides standardized training and testing partitions used for machine learning experiments on textile fibre composition estimation. The dataset consists of two HDF5 files derived from annotated samples contained in HyperTex. Each file stores hyperspectral data cubes together with their corresponding pixel-level ground-truth annotations, allowing direct integration into deep learning workflows without requiring access to the original directory structure. The partitions were created from annotated textile captures acquired using a Specim FX17 hyperspectral camera and were curated to support reproducible experimentation. The HDF5 format consolidates hyperspectral reflectance data and ground-truth composition maps into a portable representation suitable for machine learning frameworks such as PyTorch. HyperTex-Splits is intended to be used in conjunction with the original HyperTex dataset, the accompanying publication, and the official software repository, which provide complete documentation, data loading utilities, and experimental protocols.