This respository contains the main codebase for our paper: Analysis of deep learning based pose estimation techniques towards artworks classification
The notebooks uses MMPose toolbox. MMPose is an open-source toolbox for pose estimation based on PyTorch. It is a part of the OpenMMLab project.
To get it ready, follow installation instruction on https://github.com/open-mmlab/mmpose
To start experiments, you have to download historical paintings.
For our experiments, we have used the National Gallery of Art collection. The works are cataloged and it's easy to set a filter for the selected era. Thanks to the open access policy, the collection can be used in your own research.
We have downloaded 6 collections and named them accordingly: 'gothic','renaissance','baroque','rococo','classicism','romantic'. The images of each collection have been saved to the correct subdirectory in data/paintings/
- src/: The jupyter notebook code. You can find detailed description inside notebook in markdown cells.
- data/:
- json/: Calculated body proportion for each architecture.
- paintings/: Directiry where historical paintings should by downloaded.
Please consider citing if you find our findings or our repository helpful.
This work has been developed by Marcin Kutrzyński. In case of any questions or problems regarding the project or repository, do not hesitate to contact me.