MVC: A Dataset for View-Invariant Clothing Retrieval and Attribute Prediction

by Kuan-Hsien Liu, Ting-Yen Chen, and Chu-Song Chen

Citation

if you use this dataset, please cite:

Kuan-Hsien Liu, Ting-Yen Chen, and Chu-Song Chen,
"MVC: A Dataset for View-Invariant Clothing Retrieval and Attribute Prediction",
ACM International Conference on Multimedia Retrieval, ICMR 2016.

paper

Introduction

We collect the MVC dataset by crawling images from several online shopping websites, such as Amazon.com, Zappos.com or Shopbop.com. The challenge in constructing the dataset is to gather complete four different views (front, back, left, and right views) for each clothing item, as there may be only two or three views available for some clothes.

Here, we only provide 161,260 annotated images (1920 x 2240 resolutions) with 264 attribute labels. (where the total images here are different to the paper).

• image_links.mat: image links.

• attribute_labels.mat: contains all image attributes labels (264 attributes).

• mvc_info.mat: all information of 161,260 annoted images.

• json files for image_links, attribute_labels, mvc_info. (Updated: 2017/09/29)

Contact

Please feel free to contact Kuan-Hsien Liu ([email protected]), Ting-Yen Chen ([email protected]), or Chu-Song Chen ([email protected]) if you had any questions.