broadinstitute/axonet-neuromorpho-dataset

AxoNet NeuroMorpho数据集包含来自NeuroMorpho.org的神经元形态的多视角渲染图像，用于训练分割和多模态（CLIP风格）模型。数据集包含7,158个经过筛选的神经元，每个神经元从24个视角渲染，总计约164,000张图像。每个神经元包括：二值掩码（剪影）、语义分割掩码（胞体、轴突、树突）、深度图和丰富的元数据（物种、脑区、细胞类型、形态计量学）。数据集支持语义分割、深度估计、对比学习和零样本分类等任务。数据集结构包括数据清单、图像文件和元数据文件。物种分布包括小鼠、大鼠、人类等。数据集经过下载、质量控制、异常值移除和物种平衡等筛选过程。

The AxoNet NeuroMorpho Dataset contains multi-view rendered images of neuronal morphologies from NeuroMorpho.org, prepared for training segmentation and multimodal (CLIP-style) models. The dataset includes 7,158 curated neurons rendered from 24 viewpoints each, totaling approximately 164,000 images. Each neuron has: binary masks (silhouette), semantic segmentation masks (soma, axon, dendrites), depth maps, and rich metadata (species, brain region, cell type, morphometrics). The dataset supports tasks such as semantic segmentation, depth estimation, contrastive learning, and zero-shot classification. The dataset structure includes data manifests, image files, and metadata files. Species distribution includes mouse, rat, human, and others. The dataset underwent a curation process involving download, quality control filters, outlier removal, and species balancing.