A Multi-Dataset Benchmark for Generalisable Deep Learning Segmentation of Composite Microstructures Across Materials and Manufacturing Processes

This record provides a multi-domain X-ray computed tomography (XCT) composite microstructure segmentation dataset with three manufacturing domains: CFRP, FFRP-A (autoclave-cured), and FFRP-O (oven-cured). The dataset includes XCT image slices and corresponding three-class semantic segmentation masks for composite constituents/defects. Domains:- CFRP- FFRP-A (autoclave-cured)- FFRP-O (oven-cured) Labels (mask classes):- 0: matrix- 1: fibre- 2: pores/voids The dataset is released to support reproducible benchmarking of in-domain performance, cross-domain robustness, and strict out-of-domain (OOD) generalization protocols (A0–C3) as used in our accompanying manuscript.