An introduction to computer vision for working with digitised heritage collections: workshop @DHN 2020
Registration:http://dig-hum-nord.eu/conferences/dhn2020/registration/
- researchers, curators and librarians interested in using computer vision with digitised visual materials.
- experience with Python and Jupyter notebooks will be advantageous but it should be possible to follow without this.
One of the challenges in digital humanities is diversifying the types of sources that we can explore with computational methods, whether at scale or in smaller collections. In particular, digitised image collections are a rich resource for humanities research which post particular challengees for working with at scale.
Computer vision has improved greatly in the last decade thanks to deep learning approaches. For DH, this means that by leveraging ‘transfer learning’, image augmentation and other techniques, minimal training data is sufficient to train accurate models that automate image classification, illustration type detection, and similarity clustering.
This workshop will have a practical focus on getting started using computer vision tasks. Theoretical ideas will be introduced when it aids this practical work. This will include
- Introducing a pragmatic approach to training image classification models without the need for large resources
- How to build a training dataset for training new models.
- How to evaluate a classification model and understand how well it is working
Although this session won't cover working with maps in detail it will briefly provide an overview of how the techniques introduced in the session can be used for working with digitised map collections.
The workshop will partially draw on work being done as part of the Living with Machines digital history project: http://livingwithmachines.ac.uk/.