5 Key Takeaways

• 1

  The deep learning model SMARTERS reconstructs molecular geometries from simulated tip-enhanced Raman spectroscopy images.

• 2

  SMARTERS converts hyperspectral TERS data into two-dimensional maps of atomic positions, potentially reducing manual interpretation.

• 3

  The model achieved a mean Dice similarity coefficient of 0.842 for atomic-position prediction on the test set.

• 4

  Performance declined for non-planar molecules due to weaker signals from atoms farther from the scanning probe tip.

• 5

  SMARTERS struggled with experimental TERS images, failing to predict atomic positions due to differences from simulations.