Paper Doll Dataset

This dataset is proposed in the paper, (K. Yamaguchi, M. Hadi Kiapour, & T. L. Berg. 2013) Paper Doll Parsing: Retrieving Similar Styles to Parse Clothing Items..

https://github.com/kyamagu/paperdoll

Dataset Schema

• data/paperdoll_dataset.mat: 568,340 snap image urls with categories
  • 53 categories
• data/fashionista_v0.2.mat: 685 snap images and segmentation maps
  • truths (struct array): ground truth annotatation
  • predictions (struct array): predicted parsing results
  • test_index: samples used for training

See the official page's Data foramt for the original schema of the mat files.

Our json schema is the followings.

• labels/paperdoll.json
  • 568,340 samples
  • snap_id (int): starts from 1
  • snap_url (str): image url
  • post_url (str): post page url including the snap
  • items
    • category_id (int): 1 to 53
    • category (str)
• labels/categories.tsv
  • category_id (int): 1 to 53
  • category (str)

Setup

pipenv sync

or

pip install -r requirements.txt

cd $DATASET_ROOT/raw
wget http://vision.cs.stonybrook.edu/~kyamagu/paperdoll/data-v1.0.tar
tar xvf data-v1.0.tar

wget https://github.com/kyamagu/paperdoll/raw/master/data/chictopia/chictopia.sql.gz
gunzip -c chictopia.sql.gz | sqlite3 chictopia.sqlite3

python mat2json_paperdoll.py

the followings will be made from data/paperdoll_dataset.mat.

• labels/paperdoll.json
• labels/categories.tsv

python download_images.py

If time out occurs,

• increase -r (the number of requests) more than 100 (default)
• increase -t (the time to wait until time out) more than 180[s] (default)