MVC
Classification
许可协议: Unknown

Overview

This dataset not only has four different views for each clothing item, but also provides 264 attributes for describing clothing appearance. We adopt a state-of-the-art deep learning method to present baseline results for the attribute prediction and clothing retrieval performance. We also evaluate the method on a more difficult setting, cross-view exact clothing item retrieval. This dataset can be used for further studies towards view-invariant clothing retrieval.

Data Collection

We collect the MVC dataset by crawling images from several online shopping websites, such as Amazon.com, Zap- pos.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.

In current stage, the MVC dataset consists of 37,499 items and 161,638 clothing images, where most items have at least four views. Most of the image resolutions are 1920 × 2240 pixels, which thus offers sufficient details for various clothing related studies such as clothing attribute localization and type classification.

Data Annotation

To measure the clothing retrieval accuracy, we propose to use clothing attributes to establish the relevance between two images. We collect the ground truth attributes from the websites and manually select 264 attributes for similarity evaluation.

These 264 attributes are organized into a three-layer hier- archy. The first layer enforces the gender of clothes, which contains two attributes, Men and Women. There are eight categories for Men’s clothes and nine categories for Women’s clothes, where most of them overlap.

img

The third layer contains more detailed attributes. Eg., in the branch “Women->Shirts & Tops”, there are type, color, pat- tern and style attributes. Type includes Blouses, Button Up Shirts, T Shirts and Tank Tops. Color involves blue, brown, green, ... etc. Horizontal stripes, floral print are some at- tributes belonged to Pattern. Style contains attributes like short-sleeve, round-neckline, long-dress.

Citation

@inproceedings{liu2016mvc,
  title={Mvc: A dataset for view-invariant clothing retrieval and attribute prediction},
  author={Liu, Kuan-Hsien and Chen, Ting-Yen and Chen, Chu-Song},
  booktitle={Proceedings of the 2016 ACM on International Conference on Multimedia Retrieval},
  pages={313--316},
  year={2016}
}
数据概要
数据格式
Image,
数据量
161.26K
文件大小
--
发布方
Academia Sinica
Academia Sinica, the most preeminent academic institution of the Republic of China (Taiwan), was founded in China in 1928 to promote and undertake scholarly research in the sciences and humanities. After the ROC government moved to Taiwan in 1949, Academia Sinica was re-established in Taipei. Academia Sinica’s growth during this transition period was initially slow due to political instability and meager budgets.
数据集反馈
出错了
刚刚
timeout_error
立即开始构建AI
出错了
刚刚
timeout_error