How bestroi150/lumina works
Lumina is a Streamlit application for turning manuscript page images into editable transcriptions, text-mining results, geographic entities, geolocations, and downloadable outputs. Users move through five documented stages; the app also accepts text-file overrides and custom recognition models. Most stage logic resides in app.py, with Kraken operations and baseline visualization delegated to library modules. The source excerpts are partial, so stage links rely on the README’s documented workflow; detailed internal calls for unsampled stages are not inferred.
Main entry points: app.py Transcription · kraken_utils.py Baseline segmentation · blla.mlmodel BLLA model · models HTR models
Components 17
User workflow
- Results export
app.py
OCR and models
- Transcription
app.py - Baseline segmentation
lib/kraken_utils.py - Text recognition
lib/kraken_utils.py - BLLA model
data/default/blla.mlmodel - HTR models
models - Baseline visualization
lib/display_utils.py - Model metadata
lib/constants.py
Text analysis
- Data mining
app.py - Text-file override
app.py - Named entity recognition
app.py - French spaCy model
Geolocation
- Geolocation
app.py - Groq API
- Map and history
app.py
App and visualization
- Streamlit workflow
app.py
Other components
- Document analyst
Connections 17
- Document analyst → Streamlit workflow: uses
README.md - Streamlit workflow → Transcription: renders
README.md - Transcription → Baseline segmentation: segments images
app.py - Baseline segmentation → BLLA model: loads model
- Transcription → Text recognition: recognizes text
app.py - Text recognition → HTR models: uses selected model
README.md - Transcription → Baseline visualization: shows overlays
README.md - Streamlit workflow → Model metadata: uses metadata
- Transcription → Data mining: provides text
README.md - Document analyst → Text-file override: uploads text
README.md - Text-file override → Data mining: overrides input
README.md - Data mining → Named entity recognition: feeds text
README.md - Named entity recognition → French spaCy model: extracts entities
README.md - Named entity recognition → Geolocation: forwards places
README.md - Geolocation → Groq API: requests coordinates
app.py - Geolocation → Map and history: records results
README.md - Streamlit workflow → Results export: offers downloads
README.md