UmarMubiru/usl-recognition-app

Uganda Sign Language Streamlit app using 338-feature SVM model # Streamlit USL App (Hosted Setup) This package is ready for Streamlit Cloud deployment with the same UI as your current app and compressed distilled model as the default inference model. ## Folder contents - app.py - model.pkl - requirements.txt ## Input modes in app - Upload Video (best fidelity for this model) - Upload Image - Use Webcam Snapshot ## Streamlit Cloud deployment steps 1. Push this `streamlit_app` folder to GitHub. 2. Open Streamlit Cloud and sign in with GitHub. 3. Click New app. 4. Select your repository and branch. 5. Set Main file path to `streamlit_app/app.py`. 6. Click Deploy. ## Notes - Default behavior now loads `models_dataset1/csv_models/artifacts/distilled_student.joblib` first (compressed model). - If that file is unavailable, app falls back to `streamlit_app/model.pkl`, then `models_dataset1/csv_models/artifacts/best_model.joblib`. - You can still override with `MODEL_ARTIFACT_PATH` in Streamlit Cloud secrets/environment. - To deploy compressed model with safety fallback, set: - `MODEL_ARTIFACT_PATH=models_dataset1/csv_models/artifacts/distilled_student.joblib` - `FALLBACK_MODEL_ARTIFACT_PATH=models_dataset1/csv_models/artifacts/best_model.joblib` - `ENABLE_FALLBACK=true` - `FALLBACK_CONFIDENCE_THRESHOLD=0.75` If fallback is enabled and the compressed model confidence is below the threshold, the app serves the teacher prediction for that request while keeping the same interface.