ModaNet

Table of Contents

• Why we made ModaNet
• Labels
• Contributing
• Contact
• Citing ModaNet

ModaNet is a street fashion images dataset consisting of annotations related to RGB images. ModaNet provides multiple polygon annotations for each image.
Each polygon is associated with a label from 13 meta fashion categories. The annotations are based on images in the PaperDoll image set, which has only a few hundred images annotated by the superpixel-based tool. The contribution of ModaNet is to provide new and extra polygon annotations for the images.

Why we made ModaNet

ModaNet is intended to serve an educational purpose by providing a benchmark annotation set for emerging computer vision research including semantic segmentation, object detection, instance segmentation, polygon detection, and etc.

Labels

Each polygon (bounding box, segmentation mask) annotation is assigned to one of the following labels:

Label	Description
1	bag
2	belt
3	boots
4	footwear
5	outer
6	dress
7	sunglasses
8	pants
9	top
10	shorts
11	skirt
12	headwear
13	scarf & tie

The annotation data format of ModaNet follows the same style as COCO-dataset.

Data format

{
'info' : info, 'images' : [image], 'annotations' : [annotation], 'licenses' : [license],'year': year, 'categories': [category], 'type': type
}

info{
'version' : str, 'description' : str, 'contributor' : str, 'date_created' : datetime,
}

image{
'id' : int, 'width' : int, 'height' : int, 'file_name' : str, 'license' : int
}

license{
'id' : int, 'name' : str, 'url' : str,
}

annotation{
  'area': int, 
  'bbox': [x,y,width,height],
  'segmentation': [polygon],
  'image_id': int,
  'id': int,
  'category_id': int,
  'iscrowd': int
}
category{
  'supercategory': str, 'id': int, 'name': str,
}

Setup

$ git clone https://github.com/hrsma2i/modanet.git ModaNet

$ pipenv sync

or

$ pip install -r requirements.txt

Excute the following command at the parent directory of this repository (e.g. modanet/..)

$ git clone https://github.com/hrsma2i/dataset-PaperDoll.git PaperDoll

• Setup PaperDoll dataset according to the instruction in hrsma2i/dataset-PaperDoll.
  • Skip the step of downloading images.
• Caution
  • ModaNet dataset has some image ids that don't exist in PaperDoll dataset. The number of images that exist in the both, is 32,337.
  • The all urls(snap_url) in PaperDoll are not linked to main snap images that are annotated in ModaNet, so it needs to run download_images.py.

Modify .env along with you environment. Make main/labels from raw/annotations

$ pipenv shell
$ python make_main_label_from_sqlite.py
$ spltj $DATASET_ROOT/main/labels/modanet_snaps.json -s 12345
$ python make_master_category.py

Make tiny dataset.

$ smplj $DATASET_ROOT/main/labels/modanet_snaps.json -n 10 -s 12345 -o $DATASET_ROOT/tiny/labels/modanet_snaps.json
$ spltj $DATASET_ROOT/tiny/labels/modanet_snaps.json -s 12345
$ ln -sfn $DATASET_ROOT/main/labels/master_category.csv $DATASET_ROOT/tiny/labels/

Prepare images.

(new version using photos.lmdb)

$ python make_images_from_lmdb.py
$ python make_images_from_lmdb.py -s tiny

or

(old version using chictopia.sqlite3)

$ python make_url_file.py
$ cd $DATASET_ROOT/main/labels
$ aiodl -c image_urls.csv -o ../images

$ python make_url_file.py -s tiny
$ cd $DATASET_ROOT/tiny/labels
$ aiodl -c image_urls.csv -o ../images

This scripts downlaod

• main snap images in images
• their URLs in labels/image_urls.tsv

Submitting results to leaderboard

You can participate only the Object Detection task by submitting results as follows

[{
'image_id' : int, 'category_id' : int, 'bbox' : [x,y,width,height], 'score' : float,
}]

Example

[{'bbox': [192, 30, 20, 28],
  'category_id': 13,
  'image_id': 100014,
  'score': 0.8}]

You can participate only the Instance Segmentation/Semantic Segmentation/Polygon prediction tasks by submitting results as follows

[{
'image_id' : int, 'category_id' : int, 'segmentation' : polygon, 'score' : float,
}]

Example

[{'segmentation': [[210,
    31,
    212,
    35,
    204,
    37,
    204,
    45,
    205,
    54,
    199,
    58,
    194,
    52,
    198,
    42,
    192,
    32,
    194,
    30,
    201,
    33]],
  'category_id': 13,
  'image_id': 100014,
  'score': 0.8 }]

You can participate the task of joint detection and segmentation by submitting results as follows

[{
'image_id' : int, 'category_id' : int, 'segmentation' : polygon, 'score' : float, 'bbox' : [x,y,width,height]
}]

Example

[{'bbox': [192, 30, 20, 28],
  'category_id': 13,
  'image_id': 100014,
  'segmentation': [[210,
    31,
    212,
    35,
    204,
    37,
    204,
    45,
    205,
    54,
    199,
    58,
    194,
    52,
    198,
    42,
    192,
    32,
    194,
    30,
    201,
    33]],
  'score': 0.8}]

We acknowledge the contribution of COCOdataset team and all the format would follow the same style as those in the COCOdataset.

Contributing

You are more than welcome to contribute to this github repo! Either by submitting a bug report, or providing feedback about this dataset. Please open issues for specific tasks or post to the contact Google group below.

Contact

To discuss the dataset, please contact Moda-net Google Group.

Citing ModaNet

If you use ModaNet, we would appreciate reference to the following paper:

Shuai Zheng, Fan Yang, M. Hadi Kiapour, Robinson Piramuthu. ModaNet: A Large-Scale Street Fashion Dataset with Polygon Annotations. ACM Multimedia, 2018.

Biblatex entry:

@inproceedings{zheng/2018acmmm,
  author       = {Shuai Zheng and Fan Yang and M. Hadi Kiapour and Robinson Piramuthu},
  title        = {ModaNet: A Large-Scale Street Fashion Dataset with Polygon Annotations},
  booktitle    = {ACM Multimedia},
  year         = {2018},
}

License

This annotation data is released under the Creative Commons Attribution-NonCommercial license 4.0.