Synthetic and Real Image Dataset for Metrological Validation of AI-Based Circle Detection

This dataset provides both synthetic and real image data for the metrological evaluation of AI-based circle detection and droplet size estimation methods. It was created in the context of a case study on digital PCR (dPCR), where droplet radii and size distributions are relevant measurement quantities. The synthetic datasets are generated using a controlled “punch-hole” model: circular perforations are arranged on a hexagonal grid and imaged with a local mean intensity camera model. For each image, accompanying JSON metadata provides ground-truth circle centres, radii, simulation parameters, and uncertainty information derived from discretisation effects. Several dataset variants are included, covering different image resolutions, radius ranges, and Gaussian blur levels. The dataset contains real dPCR microscopy images with corresponding annotations. These data represent experimentally generated droplets and are provided in variants reflecting different preprocessing methods. Details of real data acquisition are briefly described in: Samreen Falak, Jörn Beheim-Schwarzbach, Alexander Hübner, Martin Kamme, Annemarie Martin, Hans-Peter Grunert, Ulf Dühring, Heinz Zeichhardt, Robert Ehret, Martin Obermeier, Alina Groß, Ingo Schellenberg, Andreas Kummrow, Esmeralda Valiente, Digital PCR as a potential reference measurement procedure to support monkeypox virus/Orthopoxvirus external quality assessment schemes, Methods 251 (2026) 37-44, https://doi.org/10.1016/j.ymeth.2026.03.013 The real and synthetic datasets share a compatible metadata structure, enabling their use for training, testing, benchmarking, and validation of image analysis or AI systems. The dataset is intended to support reproducible evaluation workflows for metrological AI testing, including comparison of predicted droplet radii and size distributions against reference annotations.