裁剪的VOC

从VOC2007，VOC2012的trainval集中取出car,cat,person 3类物体，用作训练集，3类共9887张；
从VOC2007的test集中取出这3类物体，用作测试集，3类共3204张。
由于有的图片包含不止1类物体，故总的图片数少于9887+3204；实际上共有11818张图。

This dataset is built by selecting images containing three target object categories (car, cat, person) from the trainval splits of VOC2007 and VOC2012 as the training set, which contains 9887 images in total for these three categories. Images of the same three categories are further selected from the test split of VOC2007 to form the test set, with 3204 images in total. Since some images encompass more than one of the target categories, the total number of unique images in this dataset is less than the sum of 9887 and 3204. In actuality, the final dataset consists of 11818 unique images overall.

搜集汇总

背景与挑战

背景概述

该数据集是从VOC2007和VOC2012中提取car、cat、person三类物体构建的，训练集包含9887张图像，测试集包含3204张图像，总共有11818张图像。数据以压缩包形式提供，包含图像文件和json标注文件，仅涵盖正样本。