MitoEM 2.0: A Benchmark of Hard Mitochondria Segmentation Datasets

MitoEM 2.0 is a collection of eight high-complexity 3D mitochondria segmentation datasets curated across diverse cell types, tissues, imaging modalities, and biological systems. The dataset focuses on biologically challenging regions characterized by dense packing, thin-neck morphologies, hyperfused networks, and ambiguous boundaries. Each dataset contains raw volumes and expert-verified ground-truth instance masks organized in nnU-Net–compatible folders. Additional descriptive files include dataset.json, split.json, difficulty metrics, and metadata.csv summarizing system, organ/tissue, cell type, voxel resolution, and imaging modality. This dataset serves as a standardized benchmark for evaluating 3D mitochondria instance segmentation algorithms.