Want to create interactive content? It’s easy in Genially!

Get started free

Manuscript Presentation

Emi Francis

Created on April 29, 2023

Start designing with a free template

Discover more than 1500 professional designs like these:

Modern Presentation

Terrazzo Presentation

Colorful Presentation

Modular Structure Presentation

Chromatic Presentation

City Presentation

News Presentation

Explore all templates

Transcript

Illuminated Manuscripts

Classification with CNN

go!

A proposal by Emi Francis

Why?

What are they?

Illuminated manuscripts are documents where text is supplemented with flourishes such as borders and miniature illustrations. They generally date from between 400 and 600 CE and into the early 16th century, but in much smaller numbers later on.

How do you classify them?

Art historians typically classify illuminated manuscripts into historic periods and types. The goal of this project is to develop a tool to help historians determine a manuscript's language or region of origin and approximate date of production. State of the art models for image classification like ResNet and CLIP can be adapted to help with this problem.

What about the data?

Many libraries and universities have digitsed manuscripts available online for free! The University of Cambridge and The Morgan Library & Museum are two institutions which maintain robust databases of thousands of manuscripts from all over the world.

Process

Prepocessing

Image Classification

OpenAI's CLIP learns by associatinng words with images, and ResNet is a staple of modern image classification. Both models should be used and the better performing one will be selected.

These documents are quite old and many are too worn to be useful for this project. It will be necessary to filter images for quality and standardize images for size.

Resources and Research

  • https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0248414
  • https://arxiv.org/abs/2107.14541
  • https://openai.com/research/clip
  • https://arxiv.org/abs/2103.00020
  • https://medium.com/@JettChenT/image-classification-with-openai-clip-3ab5f1c23e35
  • https://www.pinecone.io/learn/zero-shot-image-classification-clip/
  • https://blog.roboflow.com/how-to-use-openai-clip/

France, 15th century book of hours

Databases

  • https://www.manuscriptsonline.org/
  • https://www.cabinet.ox.ac.uk/corpus
  • https://digitizedmedievalmanuscripts.org/data
  • https://guides.lib.uchicago.edu/c.php?g=813534&p=5805534
  • https://www.themorgan.org/manuscripts/list
  • https://cudl.lib.cam.ac.uk/collections/
  • https://digital.bodleian.ox.ac.uk/browse/

Currently in Bibliothèque nationale de France