This data set is used in (X. Han et al. 2017) Bi-LSTM | Learning Fashion Compatibility with Bidirectional LSTMs..
- sequences of shop item images as an outfit
- item description
- See original dataset for details.
Install the dependencies with potry.
$ poetry install
Make raw
directory.
$ export THIS_REPO=/path/to/this/repo
$ mkdir -p $THIS_REPO/raw/labels
Download polyvore.tar.gz
and polyvore-images.tar.gz
into raw/labels
and raw
directory in this repository from the original repository.
Extract them.
$ cd $THIS_REPO/raw
$ tar zxvf $THIS_REPO/raw/labels/polyvore.tar.gz
$ tar zxvf $THIS_REPO/raw/polyvore-images.tar.gz
Move them to main
directory.
Note that main/labels
already exists,
move each file in polyvore
into main/labels
.
$ mv $THIS_REPO/raw/labels/* $THIS_REPO/main/labels
$ mv $THIS_REPO/raw/images $THIS_REPO/main
Sample data to make tiny dataset for efficient debugging.
$ brew install jq
$ cd main/labels
$ jq -c '.[]' train_no_dup.json > train_no_dup.jsonlines
$ jq -c '.[]' fill_in_blank_test.json > fill_in_blank_test.jsonlines
$ cd $THIS_REPO
$ mkdir -p tiny/labels
$ brew install coreutils
$ gshuf -n $NUM_SAMPLE main/labels/train_no_dup.jsonlines > tiny/labels/train_no_dup.jsonlines
$ gshuf -n $NUM_SAMPLE main/labels/fill_in_blank_test.jsonlines > tiny/labels/fill_in_blank_test.jsonlines
$ ./copy_images_to_tiny.sh
label/category2categorytype.tsv
: I categorised Han's category into 11 coarse categories inlabel/categorytype.tsv
.