Lacey carbon grid image dataset

This dataset contains transmission electron microscopy (TEM) images and manually prepared masks used to train and evaluate a segmentation model for identifying lacey carbon support during automated crystal searching. Version 2 extends the original dataset by providing the original image files corresponding to the processed full images used for model development and evaluation. Where available, the images are provided in their native or original acquisition formats, including H5, DM4, and TIFF files, preserving the acquisition metadata stored by the respective microscope, detector, and acquisition software. The available metadata vary between acquisition systems and individual files and may include microscope and detector information, accelerating voltage, magnification, pixel size, exposure time, image dimensions and resolution, binning, acquisition date and time, stage position, and additional acquisition parameters. The dataset includes: patches_for_training.zip – 256 × 256 pixel TEM image patches used for model training and evaluation, together with the corresponding manually annotated lacey-carbon masks. test_all_images.zip – full TEM images used to evaluate the complete crystal-detection workflow. TEM_images_of_lacey_carbon.zip – full-image dataset containing: /images – converted full TEM images used to generate the training dataset; /masks – corresponding manually prepared segmentation masks; /images_raw -- original image files corresponding to the converted images, provided in H5, DM4, and TIFF formats; image_mapping.csv, linking each converted image to its corresponding original raw file; README.txt, describing the dataset structure and the metadata available in the raw file formats. The images originated from routine crystal searching experiments conducted by different users using several microscope-detector configurations. The dataset includes different types of nanocrystalline samples and variations in image contrast and experimental conditions. Images were acquired using JEM-2100 and Zeiss Libra 120 microscopes equipped with TimePix detectors, and a Thermo Fisher Themis Z microscope equipped with a Gatan OneView detector. Raw images were converted to png and tif files for manual mask annotation and later use for model training.