Deep Learning Models for AFM Imaging: U-Net, YOLOv8, and Detectron2

This repository contains the trained deep learning models (U-Net, YOLOv8, and Detectron2) used in the article “SimuScan: Label-Free Deep Learning for Autonomous AFM”.
All models were trained exclusively on the synthetic, pre-labeled datasets generated by SimuScan, which provides diverse and physically faithful AFM-like images with built-in annotations. These datasets overcome the scarcity of training data typically required for segmentation frameworks and enable robust model training without additional manual labeling.

The deposited models demonstrate the applicability of SimuScan-generated datasets across different segmentation paradigms:

• U-Net for pixel-wise semantic segmentation.

• YOLOv8 for object detection and instance segmentation.

• Detectron2 for advanced instance segmentation.

These pretrained models serve as reference implementations for researchers interested in applying deep learning to AFM imaging, facilitating reproducibility and further exploration of label-free autonomous AFM.