民用汽车造型图像及标注数据集

本课题主要研究基于因果图的创意生成问题，旨在解决自顶而下的可控可解释数据生成问题，因此构建了一万幅级民用汽车造型图像，并高精度标注车身各组件。将产生如下科学数据：精细标注汽车图像数据集以及标注文档集。具体来说，共包含已标注汽车图像10000张，对应的标注文档保存为txt文件，共有10000份。数据共享方式拟采用完全共享方式。汽车图像数据主要来自于网上公开数据集，即Netcarshow；通过标注工具labelme部分手工标注后，构建自动打标系统进行汽车图像部件级标注，标注信息共有13种汽车部件，包括"前灯"，“尾灯”，“格栅”，“雾灯”，“侧裙”，“轮毂”，“后下护板”，”机盖”，”下格栅”，“翼子板装饰件”，“尾排”，“轮辋”，“尾翼”。数据采集地点和标注平台均为线上，采集时间贯穿整个项目执行时间（2020到2022年）。数据共享方式为数据文件下载传输。

This study focuses on creative generation based on causal graphs, aiming to solve the top-down controllable and interpretable data generation problem. Accordingly, we constructed a dataset comprising 10,000 civilian car styling images and performed high-precision annotation on each vehicle body component. The following scientific datasets will be generated: a finely annotated car image dataset and a collection of annotation documents.

Specifically, the dataset includes 10,000 annotated car images, with corresponding annotation documents saved as TXT files, totaling 10,000 in number. The planned data sharing approach is full open access.

The car image data is primarily sourced from the publicly available online dataset Netcarshow. After partial manual annotation using the annotation tool LabelMe, we developed an automatic labeling system to conduct component-level annotation for the car images. The annotation covers 13 types of automotive components, including headlights, taillights, grilles, fog lamps, side skirts, wheel hubs, rear lower guard plates, hoods, lower grilles, fender trims, exhaust pipes, wheel rims, and spoilers.

The data collection location and annotation platform are both online, and the data collection period spans the entire project execution timeframe from 2020 to 2022. The data sharing method is via downloading and transferring data files.

搜集汇总

背景与挑战

背景概述

民用汽车造型图像及标注数据集是一个包含10000张汽车图像的大规模数据集，每张图像都精细标注了13种汽车部件（如前灯、尾灯、格栅等），数据来源于公开数据集Netcarshow，并通过手工与自动结合的方式进行标注。该数据集由重庆大学创建，作为国家重点研发计划项目的一部分，旨在支持基于因果图的创意生成和计算机感知研究，数据量达5.69GB，以完全共享方式提供下载。