A Document Attention Network (DAN) model dedicated to Information Extraction from 19th century French Land Registry tables
Authors/Creators
-
1.
Laboratoire en Sciences et Technologies de l'Information Géographique pour la ville intelligente et les territoires durables
-
2.
Institut national de l'information géographique et forestière
- 3. EPITA
Description
This repository contains the weights of a Document Attention Network (DAN) (Coquenet et al.) model fine-tuned to perform extraction information from digitised tables of the 19th century French Land Registry.
The dataset used to train and evaluate the model is available on Zenodo : 10.5281/zenodo.15411507. Annotation model is described in the dataset repository.
Be aware that this model has been trained with images from the Departement of Valde-Marne. It may not suceed to extract informations from land registry documents from other departements because of other layouts, local languages, etc.
atr-dan Python library has been used to train the model.
DAN architecture as been presented in the following paper : "Denis Coquenet, Clément Chatelain, Thierry Paquet, DAN: a Segmentation-free Document Attention Network for Handwritten Document Recognition"
Files
DAN-19lr-ir-94.zip
Files
(26.2 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:bf51193079504aa8e88c2281bbf33121
|
26.2 MB | Preview Download |
Additional details
Additional titles
- Translated title (French)
- Modèle DAN affiné pour l'extraction d'informations dans les registres d'états de section du cadastre napoléonien
Related works
- Is referenced by
- Conference proceeding: 10.1007/978-3-032-05409-8_24 (DOI)
Dates
- Created
-
2025-03-12
Software
- Repository URL
- https://github.com/solenn-tl/land_registry_tables_processing_TPDL_2025
- Programming language
- Python