多场景行人检测标注图像数据

多场景行人检测数据集主要解决了一般的行人数据集中样本冗杂、场景数不足的问题。基于开源数据集WiderPerson，使用数据清洗方法剔除异常样本、清洗无效标注并最终转换为VOC数据集格式而产生。该数据集包含广泛的场景，共有行人、骑车者、部分遮挡的行人3种检测类。训练集、验证集和测试集样本数分别为1597、35和156，每个样本包含若干行人目标的信息。主要记录了图片样本中行人的位置信息、检测难易程度、行人的分类等观测值。

This multi-scenario pedestrian detection dataset primarily addresses the two core issues of redundant samples and insufficient scene diversity in conventional pedestrian datasets. Built upon the open-source dataset WiderPerson, it is developed through data cleaning procedures that remove anomalous samples, eliminate invalid annotations, and finally convert the processed dataset into the VOC dataset format. This dataset covers a wide range of scenarios and includes three detection categories: pedestrians, cyclists, and partially occluded pedestrians. The numbers of samples in the training, validation, and test sets are 1597, 35, and 156 respectively, with each sample containing information about multiple pedestrian targets. It mainly records the observed attributes such as the position information of pedestrians in the image samples, the detection difficulty level, and the pedestrian classification.

搜集汇总

背景与挑战

背景概述

多场景行人检测标注图像数据集基于WiderPerson开源数据集，经过数据清洗和转换，包含行人、骑车者和部分遮挡行人三类目标，适用于目标检测任务。数据集包含训练集、验证集和测试集，样本数分别为1597、35和156，格式为VOC数据集格式。