Retour à l'aperçu

Informations détaillées

Nom du logiciel

swissgeol-assets-dataextraction

Suite d'applications

swissgeol.ch

URL du dépôt

https://github.com/swisstopo/swissgeol-assets-dataextraction.git(publiccode.yml)

Organisation

Office fédéral de topographie (swisstopo)

Brève description

Classification pipeline for categorising PDF pages from geological reports into document classes

Description détaillée

A pipeline that classifies pages from geological PDF files into various categories such as "map" or "borehole profile", and extracts relevant metadata from the files. The output can be used to automatically generate a table of contents for each document. The pipeline is used in this way for the geological documents that are published on the web application assets.swissgeol.ch. The repository supports two classifier approaches: a feature-based XGBoost model (default) and Pixtral Large via Amazon Bedrock.

Fonctionnalités

  • Classify PDF pages into 8 geological document classes
  • Feature-based XGBoost classifier with SHAP explainability
  • Pixtral Large classifier via Amazon Bedrock
  • Versioned REST API
  • Local MLflow experiment tracking

État de développement

beta

Type de logiciel

standalone/backend

Plateformes

web

Licence

AGPL-3.0-only

Maintenance

Type de maintenance:
community
Contacts:

Localisation

Prêt pour la localisation:
Non
Langues disponibles:
en

Version Publiccode.yml

0.5.0