3DOSL - 3D Organelle Shape Library for Optical Microscopy

This dataset contains individual 3D mitochondria extracted from 3D electron microscopy images in different 3D shape representation formats of meshes, point clouds and implicit shapes. The dataset also includes synthetic microscope images of these shapes.
With more deep learning solutions being developed for fluorescence microscopy image analysis, there is increasing demand for annotated ground truth datasets for training supervised methods. However, obtaining these annotations is a laborious and expensive endeavor. To address this problem for microscope analysis of cell organelles, we release 3DOSL , a database of 3D shapes of mitochondria. 3DOSL utilizes high-resolution Electron Microscopy data as the source for creating the extensive database. Utilizing a physics-based simulator, 3DOSL allows the creation of large fluorescence microscope image datasets with 3D ground truths that can be used to train deep earning models for 3D shape reconstruction, microscope-microscope style transfer, 2D and 3D segmentation etc. We demonstrate this using a variety of example application in this paper. 3DOSL contains more than 27K instances of diverse mitochondria shapes in different 3D shape representation formats of , meshes, point clouds and implicit shapes.

本数据集汇集了源自不同三维形态表示格式——如网格、点云和隐式形状的三维电子显微镜图像中的单个线粒体。此外，该数据集亦包含这些形状的合成显微镜图像。随着深度学习解决方案在荧光显微镜图像分析领域的不断进步，对用于训练监督方法的标注真实数据集的需求日益增长。然而，获取这些标注是一项既费时又昂贵的任务。为解决细胞器显微镜分析中的这一问题，我们发布了3DOSL数据库，其中收录了线粒体的三维形状。3DOSL数据库以高分辨率电子显微镜数据为源，构建了庞大的数据库。通过采用基于物理的模拟器，3DOSL能够生成包含三维真实值的巨型荧光显微镜图像数据集，这些数据集可用于训练深度学习模型进行三维形状重建、显微镜风格迁移、二维和三维分割等。本文通过多种示例应用展示了这一过程。3DOSL包含超过27,000个不同三维形状表示格式（网格、点云和隐式形状）的线粒体实例。