MVA 2025: AI and Computer Vision for Cultural Heritage

This is the github repo for the 2025 version of the ENS MVA course on AI and Computer Vision for Cultural Heritage, taught by Prof. Dr. Robert G. Erdmann ([email protected]).

• Assignments are in the assignments folder
• Notebooks, for example those developed in class, are in the notebooks folder
• Data are, where possible, in the data folder, but in cases in which the files are too large, they may also appear in the releases section.

News

2025-03-20

• Assignment 3 is posted. I gave explicit instructions for most all of the steps, so it should be relatively straightforward. It's due on March 27.

2025-03-14

• Some additional links are posted below for some lecture notes and other useful information.
• A new notebook has been added for download of a 14k image dataset of posters from the BnF.
• A painting radiograph corresponding to Rijksmuseum object number SK-C-207 (A waterfall, Adam Pijnacker, 1649 - 1673) is posted in the data section, as a demonstration for performing canvas analysis.

2025-03-06

• Assignment 2 is posted. It is due on March 17.

Links

• Color Theory

  • PDF on basics of color theory

• The Structure Tensor

  • Lecture notes from Linköping University Provides a variety of motivations for the structure tensor and coherence measures.

• International Image Interoperability Framework (IIIF)

  • https://iiif.io/

Papers

• RANSAC-Flow:

  • Title: RANSAC-Flow: generic two-stage image alignment
  • Link: arXiv:2004.01526

• W-Net:

  • Title: The Little W-Net That Could: State-of-the-Art Retinal Vessel Segmentation with Minimalistic Models
  • Link: arXiv:2009.01907

• U-Net:

  • Title: U-Net: Convolutional Networks for Biomedical Image Segmentation
  • Link: arXiv:1505.04597

• Image Style Transfer:

  • Title: Image Style Transfer Using Convolutional Neural Networks
  • Link: CVPR Paper

• The Structure Tensor

  • Multiscale Texture Enhancement Gives explicit derivations for the eigenvalues and eigenvectors of the structure tensor in Eqs. (10) and (11).

• ConvNeXt:

  • Title: A ConvNet for the 2020s
  • Link: arXiv:2201.03545
  • Also see: GitHub Code Release

• ConvNeXt V2:

  • Title: ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders
  • Link: arXiv:2301.00808

• ResNet50:

  • Title: Deep Residual Learning for Image Recognition
  • Link: arXiv:1512.03385
  • Also see: Hugging Face Model Hub

• UMAP:

  • Title: UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
  • Link: arXiv:1802.03426
  • Also see these alternative links: GitHub, Journal of Open Source Software

• t-SNE:

  • Title: Visualizing Data using t-SNE
  • Link: JMLR Paper
  • Also See: t-SNE Homepage (Laurens van der Maaten)

• CLIP:

  • Title: Learning Transferable Visual Models From Natural Language Supervision
  • Link: arXiv:2103.00020
  • Also See: Official OpenAI CLIP GitHub, OpenCLIP (open-source implementation)

• DINOv2:

  • Title: DINOv2: Learning Robust Visual Features without Supervision
  • Link: arXiv:2304.07193
  • Also See: GitHub, HuggingFace, and Meta AI Demos

• Pigment Mapping:

  • Title: Use of imaging spectroscopy, fiber optic reflectance spectroscopy, and X-ray fluorescence to map and identify pigments in illuminated manuscripts
  • Link: Taylor & Francis Online

Online Interactive Visualizations

• R.G. Erdmann. “Automatically colorized radiograph for Vermeer’s Milkmaid, (Rijksmuseum SK-A-2344)”. 2023-01-08. Online at http://tinyurl.com/2pp9yd4s
• R.G. Erdmann. “Evidence of stress-induced craquelure through high-resolution image coherence angle analysis of Vermeer’s Milkmaid (Rijksmuseum SK-A-2344). images.erdmann.io, 2023-01-08. Online at http://tinyurl.com/r7w6vrru
• R.G. Erdmann. “Interactive visualization of canvas support thread angles in Vermeer’s Milkmaid (Rijksmuseum SK-A-2344)”, images.erdmann.io, 2023-01-10. Online at http://tinyurl.com/bdda4mvr
• R.G. Erdmann with RIS data collected by F. Gabrieli. “An online interactive tool for false-color image creation from a fused RIS-VNIR+RIS-SWIR dataset for Vermeer’s Milkmaid (Rijksmuseum SK-A-2344)”, images.erdmann.io, 2023-01-10. Online at http://tinyurl.com/4jrbjc8c
• R.G. Erdmann with MA-XRF data collected by A. van Loon. “Demonstration of a neural network-based approach for denoising and deconvolution of raw MA-XRF scans with several elemental maps from Vermeer’s Milkmaid (Rijksmuseum SK-A-2344)”, images.erdmann.io, 2023-01-12, Online at http://tinyurl.com/4t5dnm9c
• R.G. Erdmann. "Images from Rembrandt's drawing of Saskia van Uylenburgh (Rijksmuseum, RP-T-1930-51)", images.erdmann.io. Online at https://images.erdmann.io/Draper/?collection=/Rijksmuseum/RP-T-1930-51