A dataset for 2D pose estimation of anime/manga images. This repository contains code to:
bundler
(for installing the following three packages)rake
(for automation) nokogiri
(for HTML processing)mechanize
(for interaction with web pages)convert
command from the shell.First, please clone the dataset into a directory of your choice. Then, change into the directory of the repository. At this point, you should have (1) the Ruby language, (2) the bundler
package, and (3) ImageMagick installed in your system.
The next step is to install other Ruby packages. Run:
bundle install
Then, run:
rake build
The above command will download all the images and process them. This can take some time, so sit back, relax, and wait.
After the above command finishes, you can browse the dataset by viewing the HTML page index.html
in the root of the repository.
The images are located in the data/images
directory. The joint positions are located in the following files:
data/data.json
contains all the 2,000 examples.data/train.json
contains the 1,400 training examples.data/val.json
contains all the 100 validation examples.data/test.json
contains all the 500 test examples.Some joint names are a little counter-intuitive:
arm_left
and arm_right
are the shoulder joints.leg_left
and leg_right
are the hip joints.tiptoe_left
and tiptoe_right
are the tips of the shoes or feet (if the character does not wear shoes).For those who do not want to figure out how to install the dependencies, I have prepared a Docker image with all of them. The name of the image is:
dragonmeteor/animedrawingsdataset
You can also inspect the image on the web here.
To use the image, you first need to clone the repository to a location. Let us say to /opt/AnimeDrawingsDataset
. Then, you can run the image, mounting the repository directory as a data volume.
docker run -ti --net=host -v /opt/AnimeDrawingsDataset:/AnimeDrawingsDataset dragonmeteor/animedrawingsdataset /bin/bash
In the command above, we mount it to the /AnimeDrawingsDataset
in the container. Next, change to the directory and start the downloading process:
cd /AnimeDrawingsDataset
rake build
This dataset would not have been possible without volunteers who helped us annotated the pictures. Thank you very much!