SEM Nanoparticle Instance Segmentation Reference Data: Expert-Verified and Single-Pass Annotations of 295 Scans

Instance-level segmentation masks for 295 scanning electron microscopy (SEM) images of nanoparticles, comprising 46,849 annotated particle instances, including dense, agglomerated, and overlapping configurations. The images are taken from the nanoparticle subset of the SEM collection of Aversa et al. (2018). The dataset is organized in two quality tiers: - Expert-verified (153 scans, 19,052 instances): independent, mutually blinded double annotation, expert adjudication, and consensus review; scans without consensus were excluded.- Single-pass (142 scans, 27,797 instances): model predictions corrected in one manual pass by one annotator, without review. Images are 1024 × 768 px JPEGs. Masks are 16-bit single-channel PNGs with one unique integer ID per instance (0 = background). metadata.csv lists tier, construction stage, model assistance, and instance count for each scan. The dataset has no fixed benchmark split; evaluation data should be taken from the expert-verified tier. See README.md for annotation conventions, quality control, intended use, and limitations. Files: README.md, LICENSE, metadata.csv, data.zip (images and masks; extract next to metadata.csv). Contact: [e-mail]