LSC2TEXT

Details

For details on the project and my motivation on doing this (though it is only available in Spanish, for the moment): PROYECTO

How to Set Up this Project?

1. Requirements

You need to download the following:

• uv (Project and package manager)

2. Set Up the Data

1. Save the LSC70 (Available in https://data.mendeley.com/datasets/9ssyn8tff5/2) in the /data/raw folder.
2. Execute the CSV generation script by executing uv run .\scripts\generate_dataset_csv.py.
3. Execute the filtering script by executing uv run .\scripts\filter_dataset.py.
4. Split the filtered dataset into train/valid: uv run .\scripts\split_dataset.py.

This can be automatically done with default values through the shell with setup.sh.

3. Train the Model

Train the model by using uv run .\scripts\train.py.

4. Start the Backend

Start the local API server:

uv run python -m src.api.api --model-path models/registry/svm/direct-svm-20260511-220859/model.joblib --port 8000

Available endpoints:

• GET /health returns readiness information
• GET /metadata returns the service description and default model path
• POST /predict accepts one uploaded image and returns ranked predictions as JSON

The API uses the same inference pipeline as scripts/infer.py and is intended for local development and testing.

5. Access the Frontend

Visit http://127.0.0.1:8000/ after (4) to upload a single image and view the rendered prediction result.

The frontend uses server-rendered templates and is designed to work locally with the backend API.

Testing

Test can be run with uv run python -m unittest discover -s tests -p "test*.py".