TreeOfLife-10M 生物图像数据集

TreeOfLife-10M 拥有超过 1,000 万张图像，涵盖生命之树中的 45.4 万个分类单元，是迄今为止最大的 ML 就绪生物有机体图像及其相关分类标签的数据集。它扩展了现有高质量数据集（如 iNat21 和 BIOSCAN-1M）建立的基础，进一步整合了来自生命百科全书 (eol.org) 的新精选图像，这些图像提供了 TreeOfLife-10M 的大部分数据多样性。 TreeOfLife-10M 中的每张图像都标记为最具体的分类级别，以及生命之树中更高的分类等级（有关分类等级和标签的示例，请参阅文本类型）。 TreeOfLife-10M 的生成目的是训练 BioCLIP 和未来的生物学基础模型。

TreeOfLife-10M, which contains over 10 million images spanning 454,000 taxa in the Tree of Life, is the largest ML-ready dataset of biological organism images and their associated taxonomic labels to date. Building upon the foundations laid by existing high-quality datasets such as iNat21 and BIOSCAN-1M, it further integrates newly curated images from the Encyclopedia of Life (eol.org), which contribute to most of the data diversity of TreeOfLife-10M. Each image in TreeOfLife-10M is annotated with its most specific taxonomic rank, as well as higher taxonomic ranks within the Tree of Life (for examples of taxonomic ranks and labels, please refer to the text type). TreeOfLife-10M was developed specifically for training BioCLIP and future foundational biological models.

搜集汇总

背景与挑战

背景概述

TreeOfLife-10M是迄今为止最大的生物有机体图像数据集，包含超过1,000万张图像，涵盖45.4万个分类单元，每张图像都标注了详细的分类等级。该数据集旨在支持生物多样性研究、物种识别和计算机视觉任务，特别是为BioCLIP等生物学基础模型的训练提供数据支持。