A few shot learning approach to segmentation of STM scans. Corresponding paper with more details: "Overcoming_Labelled_Data_Scarcity_for_Defect_Classification_in_Scanning_Tunnelling_Microscopy".
Works for small scale (~100nm) identification of lattice defects and their classification. For Si(001), Si(001)-H, Ge(001), and TiO2(xxx)
A working progress!
FSL_STM_UI contains a script which contains a very simple user interface for the use of this algorithm.
Accuracy_variability_study contains the scripts used to assess the variability of our results due to random initialisation and validation/test set splits.