yuyan-chen/open-insect

Open-Insect数据集旨在为生物多样性监测中的新颖物种的开集识别提供一个基准。它包含了指向[全球生物多样性信息设施（GBIF）](https://www.gbif.org/)编译的图像文件的URL，以及用于训练分类器的其他元数据，如纬度、经度、GBIF物种键和标签。该数据集部分基于[AMI数据集](https://zenodo.org/records/11358689)的子集，并包含了三个区域（东北美洲、西欧和中美洲）的内分布（ID）、外分布（OOD）和辅助数据（AUX）。对于中美洲区域，还包括了从巴罗科罗拉多岛（BCI）最近收集的197张图像，涵盖133个物种，其中59个可能尚未被科学描述。

The Open-Insect dataset is curated to benchmark open-set recognition of novel species in biodiversity monitoring, focusing on insects. It consists of URLs pointing to image files compiled from the [Global Biodiversity Information Facility (GBIF)](https://www.gbif.org/) along with other metadata such as latitude, longitude, GBIF speciesKey, and label used to train the classifier. The dataset partially builds on a subset of [the AMI dataset](https://zenodo.org/records/11358689) and includes in-distribution (ID), out-of-distribution (OOD), and auxiliary data (AUX) for three regions: Northeastern North America (NE-America), Western Europe (W-Europe), and Central America (C-America). For C-America, it also includes 197 recently collected images from Barro Colorado Island (BCI), Panama, covering 133 species, 59 of which are likely undescribed by science.