Aesthetic evaluation of vehicle appearance design is an important part of the vehicle design process. How to conduct a high-confidence, rapid, objective and systematic evaluation of a given vehicle appearance is of great significance to vehicle designers, consumers, decision makers and other groups. Big data-driven deep learning approach to automated, reliable, multi-dimensional and systematic evaluation of vehicle appearance relies on high-quality datasets. There is a lack of datasets for exterior styling of automobiles. And the previous methods have the subjective problems of results, limitation of evaluation dimensions, and lower consumer involvement. Aiming at the above problems, this paper combines intelligent mechanisms such as quantitative scoring and perceptual imagery evaluation of vehicle appearance from the perspectives of vehicle users, styling experts, etc., and the main research contents are as follows: Create a large-scale and multi-dimensional vehicle appearance datasets (VAD), with near 140000 images, which is merged by three datasets: MVVA, VAS and VAPIE. Multi-view vehicle appearance image dataset (MVVA) contains 2093 automobile models and 78590 images, with small samples in the aesthetic evaluation of vehicle appearance design; Vehicle appearance scoring data set (VAS) with 19610 images is collected, cleaned and preprocessed online then labeled by the quantitative scoring and perceptual imagery evaluation of the two dimensions of users and experts; And vehicle appearance perceptual imagery evaluation dataset (VAPIE) with 40017 images based on the perspectives of the users and the experts is labeled from the image data. Our dataset is now available on Github: https://github.com/KDafu/Vehicle-Appearance-Dataset.