SEM-EDS MXene Segmentation

Semantic segmentation of SEM images of MXene tiles into four classes (BG, Flake, Circle, Other) with a UNet++ (ResNet-18 encoder), followed by SEM-EDS/EDX composition analysis of the segmented morphologies.

The repository ships a small annotated dataset and demonstrates that a copy-paste augmentation step improves a low-data baseline. Every experiment is run over three seeds (1, 52, 88) and results are reported as mean ± s.d. across the three runs. All scripts resolve their paths relative to the repository root, so the project runs from anywhere after cloning.