TsinghuaDog

清华狗是一个细粒度的狗分类数据集，超过65% 的图像是从人们的现实生活中收集的。数据集中的每个犬种至少包含200张图像，最多包含7,449张图像，基本上与它们在中国的出现频率成比例，因此与现有数据集相比，它显着增加了每个犬种的多样性。

Tsinghua Dogs is a fine-grained dog classification dataset. Over 65% of its images were collected from real-life scenarios. Each dog breed in the dataset contains no less than 200 images and up to 7,449 images, with the number of images roughly proportional to the actual occurrence frequency of the breed in China. Consequently, compared with existing datasets, it significantly increases the diversity of each dog breed.

搜集汇总

背景与挑战

背景概述

TsinghuaDog是一个细粒度的狗分类数据集，包含大量现实生活中的犬种图像，每个犬种的图像数量与其在中国的出现频率成比例。该数据集由清华大学、哈佛医学院和纳九公司联合发布，适用于图像分类任务。