This page is for contributors working on the repo itself rather than just running a single experiment.
| Area | When you touch it |
|---|---|
datasets/ |
ingestion, manifests, slide extraction, dataset utilities |
models/ |
PyTorch models, training code, conversion code |
scripts/ |
orchestration wrappers and pipeline helpers |
benchmarks/ |
benchmark entry points and result handling |
tools/ |
desktop runner, reporting, comparison, and analysis utilities |
android/ |
mobile app, workers, TFLite inference, and UI |
python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements.txtOptional packages by workflow:
python -m pip install -r requirements-optional.txt- system
ffmpegfor video extraction workflows - system Tesseract OCR if you use OCR evaluation
cd android
.\gradlew tasks
.\gradlew assembleDebugOpen android/ in Android Studio if you need:
- SDK setup
- emulator/device debugging
- Compose previews
- logcat inspection
pytest tests/unit -q
python -m models.training.finetune_education --help
python -m models.training.evaluation --help
python tools/benchmark-runner/runner.py --help- Validate manifests before long training runs.
- Keep benchmark outputs out of source folders unless they are intentionally checked in.
- When changing Android processing behavior, review the docs in
docs/architecture/anddocs/api/as part of the same pass.