Crowd Activity

该数据集集中于人群的活动，以进行细粒度的图像分类任务，称为人群活动数据集，因为自动了解人群活动对社会安全很有意义。这个数据集是新收集的，其中图像主要在互联网上搜索，并通过手机从街道上收集。此数据集中的所有图像至少包含一个文本实例。这些类别来自近年来受热点事件刺激的日常生活活动和示威活动。具体来说，该数据集总共包含21个类别和8785个图像。这21类大致分为两种: 日常生活活动 (即庆祝圣诞节、举办运动会、举办音乐会、庆祝生日派对、名人演讲、教学、毕业典礼、野餐、新闻发布会、购物、庆祝感谢日) 和示威活动 (即保护动物，保护环境，呼吁和平，英国退欧，新型冠状病毒肺炎，选举，移民，尊重女性，种族平等，贾奇运动)。

This dataset focuses on crowd activities for fine-grained image classification tasks, and is thus named the Crowd Activity Dataset, as automatic understanding of crowd activities is of great significance for social security. This newly collected dataset primarily comprises images sourced from Internet searches and street captures via mobile phones. All images in this dataset contain at least one text instance.
The category system is derived from daily activities and demonstrations stimulated by recent hot events. Specifically, this dataset encompasses a total of 21 categories and 8785 images. The 21 categories are roughly divided into two groups: daily activities and demonstrations.
Daily activity categories include: celebrating Christmas, holding sports events, holding concerts, celebrating birthday parties, celebrity speeches, teaching sessions, graduation ceremonies, picnics, press conferences, shopping, and celebrating Thanksgiving Day.
Demonstration categories include: animal protection, environmental protection, calls for peace, Brexit, COVID-19 related protests, elections, immigration-related activities, respect for women, racial equality, and the Jiaqi Movement.

搜集汇总

背景与挑战

背景概述

Crowd Activity是一个细粒度图像分类数据集，包含21个类别共8785张图像，分为日常生活活动和示威活动两大类，所有图像至少包含一个文本实例。该数据集由华中科技大学和腾讯优图于2022年发布，旨在支持社会安全相关的自动人群活动理解研究。