jeffbbrown2/ConnectomeBench2

ConnectomeBench2是一个统一的基准数据集，用于连接组神经分割数据的自动校对。它包含401,170个样本，覆盖4个物种（小鼠、果蝇、人类、斑马鱼）和5种样本类型（真实合并编辑、真实分割编辑，以及合成的相邻/连接/突触控制），并附带相关的网格几何和电子显微镜（EM）渲染数据。ConnectomeBench2 is a unified benchmark for automated proofreading of connectomic neural-segmentation data. 401,170 samples across 4 species (mouse, fly, human, zebrafish) and 5 sample types (real merge edits, real split edits, synthetic adjacent / junction / synapse controls), with the associated mesh geometry and electron-microscopy (EM) renderings.