SELMA3D2024

在这个任务中，每个参与者将收到一个包含两组的训练数据集。第一组包括大量（> 6x10^11 体素，相当于 > 35000 张 256x256x256 体素的图像）没有注释的小鼠和人类样本的全脑 3D LSM 图像，该组应该有助于通过自监督学习进行模型预训练。第二组由带有精确注释的全脑 3D LSM 图像的裁剪补丁组成，可以对模型进行微调以执行语义分割任务。In this task, each participant will receive a training dataset consisting of two subsets. The first subset includes a large-scale collection (> 6×10¹¹ voxels, equivalent to over 35,000 256×256×256 voxel images) of unannotated whole-brain 3D LSM images from mouse and human samples, which is intended to support model pre-training via self-supervised learning. The second subset comprises cropped patches of whole-brain 3D LSM images with precise annotations, which can be used to fine-tune models for semantic segmentation tasks.

搜集汇总

背景与挑战

背景概述

SELMA3D2024数据集是SELMA3D挑战赛的训练集，专门用于3D光片显微镜图像的自监督学习分割任务。该数据集包含大脑中血管、c-Fos标记的脑细胞、细胞核和阿尔茨海默病斑块等生物结构的3D图像块，每个块均配有像素级注释，数据来源于小鼠和人类大脑样本，采用CC BY 4.0许可。