Description

At a time where we benefit from big collections of digitized material such as paintings and drawings, technological tools have a great potential to enhance our interactions and allow new researches with these collections.

However, most machine learning algorithms were and still are trained with real images, pictures gathered from online applications such as Flickr and which content mostly represent daily scenes of our western society. Such models, when applied on other types of images such as artworks, have lower accuracy and have to be re-trained. Unfortunately, there is a lack of proper training datasets for artworks, especially for detection tasks such human pose estimation (HPE). It is because of this issue that the Pose Annotation Project for Artworks (PAPA) was created.

The goal of the project is to create a platform where people can annotate paintings and drawings. These annotations will follow standard format for common machine learning tasks so that the training of existing models is simplified.

Details

The annotation process will be divided into three following tasks in order to ease the operations for the annotators :

1. Segmentation
2. Keypoints annotation
3. Correction

News

The web application for the annotation process will be online in early 2022 and further details will be provided soon.