真菌分类数据集

1. 简介FungiTastic 是围绕野生真菌观察构建的大规模多模态基准数据集，旨在为机器学习和生物多样性研究提供资源；包含约 350,000 次观测、超 600,000 张图像，覆盖超 5000 个物种，数据经专家标记（部分经 DNA 测序验证），收集时间超 20 年；分为三个子集，分别是含所有模态、超 346k 观测和 4500 + 物种的完整子集，含 6 个属、70k 图像及身体部位分割掩码的迷你子集（FungiTastic-M），以及含 2. 文件类型包含多种类型文件，视觉类有真菌照片、孢子显微图像、身体部位分割掩码文件；文本类有图像标题文件、观测元数据文件；地理气候类有地理空间数据文件、气候时间序列数据文件，且所有文件按不同子集和任务需求分类组织

1. Introduction
FungiTastic is a large-scale multimodal benchmark dataset constructed around wild fungal observations, intended to serve as a valuable resource for machine learning and biodiversity research. It encompasses approximately 350,000 observations and over 600,000 images, covering more than 5,000 species. The data are expert-annotated, with a portion verified through DNA sequencing, and the data collection process has spanned over 20 years.
The dataset is divided into three subsets:
1. The Full Subset: includes all modalities, with over 346,000 observations and more than 4,500 species;
2. The Mini Subset (FungiTastic-M): covers 6 genera, 70,000 images, and body part segmentation masks;
3. The dataset includes multiple types of files, categorized as:
- Visual files: fungal photographs, microscopic spore images, and body part segmentation mask files;
- Text files: image caption files and observation metadata files;
- Geospatial and climatic files: geospatial data files and climatic time-series data files.
All files are organized and classified according to different subsets and task requirements.

搜集汇总

背景与挑战

背景概述

FungiTastic是一个大规模多模态真菌分类数据集，包含约350,000次观测和超600,000张图像，覆盖超5,000个物种，适用于细粒度图像分类、少样本学习及多模态学习等多种计算机视觉和生物多样性研究任务。