A Multi-Microscope Annotated Dataset of Real Microscopy Images of Microorganisms for Classification, Detection, and Segmentation
Year: 2026
Type: dataset
Abstract: This dataset contains 100,000 microscopic images of microorganisms belonging to five classes (micrococci, diplococci, streptococci, bacilli, and backgrounds), along with corresponding annotations in the form of bounding boxes and binary segmentation masks. The dataset is designed for the training, validation, and comparative testing of computer vision models aimed at analyzing microbiological images acquired under diverse conditions. Dataset structure The dataset is organized into two folders: images/ — microscopy images in PNG formatmasks/ — corresponding binary segmentation masks in PNG format Each image and its mask share the same filename. On the masks, white pixels with a value of 255 correspond to foreground regions, whereas black pixels with a value of 0 represent the background. The dataset is specifically constructed to reflect the variability encountered in real-world diagnostic laboratories. During data acquisition, several microscopes from the same Levenhuk MED series were used interchangeably within the experimental setup, including the Levenhuk MED 20T, MED 25T, MED 30T, MED 35T, and MED 40T. Thus, the acquisition system was not restricted to a single microscope: microscopes could be replaced within the setup while maintaining the same general acquisition and automation workflow. In addition to microscope variability, the dataset includes variations in magnification settings, illumination configurations, staining procedures, spatial resolution, object scale, brightness, contrast, sharpness, and color distribution. This heterogeneity ensures that models trained on this dataset are exposed to a broad range of microscopy acquisition conditions. The dataset exhibits substantial statistical variability across the five microorganism classes. Depending on the class, image resolutions range approximately from 40 × 40 to 759 × 661 pixels. Mean grayscale brightness ranges from approximately 173 to 178 intensity levels, while mean grayscale contrast varies between 8.3 and 10.8. The mean dynamic range ranges from approximately 55 to 91, and the average image entropy varies between approximately 4.4 and 5.0 bits. Laplacian variance, used as an indicator of image sharpness and focus quality, ranges from approximately 35 to 89, reflecting considerable variation in focus and fine-detail preservation. Mean image colorfulness ranges from approximately 27.1 to 31.1, indicating moderate chromatic variability across the acquired microscopy images. The annotations also contain substantial geometric variability. The average foreground fraction ranges from approximately 26% for bacilli to 45% for diplococci and streptococci, providing images with substantially different levels of microorganism coverage. Bounding boxes are generally centered within the extracted image regions, with normalized center coordinates close to 0.50, while bounding-box aspect ratios span a broad range of approximately 0.14-7.26, reflecting the morphological diversity of the represented microorganism classes. Number of image/mask pairs100,000 Images per class20,000 Number of classes5 Image resolution rangeapproximately 40 × 40 to 759 × 661 pixels Mean grayscale brightnessapproximately 173-178 Mean grayscale contrastapproximately 8.3-10.8 Mean dynamic rangeapproximately 55-91 Mean image entropyapproximately 4.4-5.0 bits Laplacian varianceapproximately 35-89 Mean colorfulnessapproximately 27.1-31.1 Foreground fractionapproximately 26-45% Bounding-box aspect ratio rangeapproximately 0.14-7.26 Microscopes usedLevenhuk MED 20T, MED 25T, MED 30T, MED 35T, MED 40T Image formatJPG, PNG Mask formatPNG
Subfield: Biophysics
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7172200470
Cites:
Cited by:
Related to: