Dataset for AI-assisted Atomic-Resolution STEM Imaging and Trajectory Analysis

This dataset contains experimental and synthetic data used in the study“Probing hidden materials interactions through machine-learned atomic-resolution electron microscopy”. The data support the development and validation of DAI2SY (Deep Atomic-resolution Imaging and AI Sensing with YOLO),a deep learning framework for denoising, recognizing, and tracking single-atom dynamics in low-dose in situHAADF-STEM image sequences. The dataset includes:(1) Raw and processed HAADF-STEM image sequences of Pt single atoms supported on graphene and amorphous carbon,acquired using aberration-corrected STEM at 60 kV;(2) Synthetic atomic-resolution STEM images generated by multislice simulations and combined with experimentallyextracted noise for synthetic-to-real transfer learning;(3) Denoised image outputs, atomic coordinate data, and reconstructed atomic trajectories used for statisticalanalysis of atomic motion modes, including localized vibration and directional hopping. Due to the large size of the dataset, it is hosted on Zenodo and is referenced by the associated code repository.