Scanning Electron Microscopy (SEM) Dataset of Additively Manufactured Ni-WC Metal Matrix Composites for Semantic Segmentation

This dataset accompanies the publication “Accelerated quantification of reinforcement degradation in additively manufactured Ni-WC metal matrix composites via SEM and vision transformers.” It contains scanning electron microscopy (SEM) images and corresponding pixel-level segmentation masks used to train, validate, and test deep learning models for microstructural analysis.

The images represent metallographic cross-sections of directed energy deposited (DED) nickel-tungsten carbide (Ni-WC) metal matrix composites (MMCs), prepared through a detailed process involving sectioning, polishing, and etching. Each SEM image highlights key microstructural constituents — matrix, carbide particles, dilution bands, and reprecipitated carbides — with the latter two representing degradation features of reinforcement particles under elevated thermal conditions.

The dataset includes:

• AugmentedImages.zip: 405 SEM image crops (512×512 pixels each) obtained from four magnification levels (1000×, 800×, 700×, 600×) and augmented via flips, rotations, elastic transforms, grid distortions, and controlled contrast/brightness variations.

• AugmentedMasks.zip: Corresponding manually labeled segmentation masks, generated using the Supervisely interface with pixel-level accuracy and verified for class consistency.

The data were generated and augmented using the Albumentations library to enhance generalization and support high-throughput segmentation studies. The dataset enables benchmarking and comparison of convolutional (e.g., DeepLabV3+) and transformer-based architectures (e.g., SegFormer, UPerNet, Mask2Former, etc.) for microstructure segmentation in additive manufacturing contexts.

Applications:

• Training and evaluation of semantic segmentation models for materials characterization

• Quantification of carbide degradation and process-induced defects in Ni-WC MMCs

• Benchmarking of vision transformer architectures on SEM datasets

File format:

• Images: .bmp

• Masks: .bmp

• Resolution: 512×512 pixels

Acknowledgment:
If you use this dataset, please cite the associated article and acknowledge the dataset authors.