iNaturalist 2017

iNaturalist 2017数据集 (iNat) 包含来自5,089自然细粒度类别的675,170训练和验证图像。这些类别属于13个超级类别，包括植物 (植物)，昆虫 (昆虫)，鸟类 (鸟类)，哺乳动物 (哺乳动物) 等。iNat数据集高度不平衡，每个类别的图像数量截然不同。例如，最大的超级类别 “Plantae (植物)” 具有来自2,101类别的196,613图像; 而最小的超级类别 “Protozoa” 仅具有来自4个类别的381图像。iNaturalist 2017数据集 (iNat) 包含来自5,089自然细粒度类别的675,170训练和验证图像。这些类别属于13个超级类别，包括植物 (植物)，昆虫 (昆虫)，鸟类 (鸟类)，哺乳动物 (哺乳动物) 等。iNat数据集高度不平衡，每个类别的图像数量截然不同。例如，最大的超级类别 “Plantae (植物)” 具有来自2,101类别的196,613图像; 而最小的超级类别 “Protozoa” 仅具有来自4个类别的381图像。

The iNaturalist 2017 dataset (iNat) comprises 675,170 training and validation images across 5,089 fine-grained natural categories. These categories are grouped into 13 supercategories, including plants, insects, birds, mammals, and others. The iNat dataset is highly imbalanced, with substantial variations in image counts per category. For instance, the largest supercategory, *Plantae* (plants), contains 196,613 images across 2,101 categories; whereas the smallest supercategory, *Protozoa*, only includes 381 images from 4 categories. The iNaturalist 2017 dataset (iNat) comprises 675,170 training and validation images across 5,089 fine-grained natural categories. These categories are grouped into 13 supercategories, including plants, insects, birds, mammals, and others. The iNat dataset is highly imbalanced, with substantial variations in image counts per category. For instance, the largest supercategory, *Plantae* (plants), contains 196,613 images across 2,101 categories; whereas the smallest supercategory, *Protozoa*, only includes 381 images from 4 categories.

AI搜集汇总

背景与挑战

背景概述

iNaturalist 2017是一个大规模细粒度图像分类数据集，包含675,170张训练和验证图像，覆盖5,089个自然类别，分为13个超级类别如植物、昆虫和鸟类。该数据集高度不平衡，不同类别的图像数量差异极大，例如最大类别'Plantae'有超过19万张图像，而最小类别'Protozoa'仅381张，适用于研究不平衡数据下的分类任务。