AFM Image Dataset: Synthetic and Real, Labeled and Unlabeled

This dataset contains grayscale PNG images of bacterial and geometric shapes acquired using Atomic Force Microscopy (AFM), as well as procedurally generated synthetic images for machine learning applications in detection and segmentation. Real AFM images were collected using the **DriveAFM system (Nanosurf)** equipped with **Multi75 probes**, under standard tapping mode conditions. The samples include both **bacterial morphologies** (bacillus and coccus) and **geometric test shapes**, and are provided in labeled and unlabeled formats. Synthetic datasets were generated to mimic realistic AFM topography using domain-specific simulators, incorporating noise, tip convolution, shape variation, and texture. The synthetic labeled images include pixel-level annotations compatible with COCO and YOLO formats. The dataset is structured into multiple folders:- `synthetic_labeled/` — synthetic images with segmentation masks- `real_labeled/` — real AFM images with manual annotations- `real_unlabeled/` — real images without labels All files are grayscale `.png` images, 8-bit depth. Annotations (when present) are included in `.json` format. These datasets can be used for training, validation, benchmarking, and domain adaptation in AFM image analysis tasks. This dataset was used in the publication: > **[Insert title of the article]** > DOI: [Insert article DOI] Please cite the above article when using this dataset. **License**: Creative Commons Attribution 4.0 International (CC BY 4.0) **Author**: Ruben Millan-Solsona **Version**: 1.0 **Date**: August 2025