MALLI

MALLI is a blood-smear malaria analysis project that combines a TensorFlow training and evaluation pipeline with a Flutter mobile app for offline-friendly sample capture and result review.

What the project does

The repository contains two connected parts:

• A Python ML stack for loading malaria datasets, training a MobileNetV3-small classifier, evaluating saved weights, and exporting INT8 TFLite models.
• A Flutter app that displays completed samples, captures new images, and stores results locally with SQLite.

The main Python entry points are train.py, models/evaluate.py, and the helpers under models/. The Flutter app starts from lib/main.dart.

Why the project is useful

MALLI is designed for field and low-resource workflows where malaria screening needs to be portable, repeatable, and easy to review. It provides:

• A staged training pipeline for NIH and synthetic blood-smear data.
• TFRecord caching for faster dataset reuse on larger runs.
• Quantized model export for mobile deployment.
• A cell-counting pipeline for parasite and ROI analysis.
• A simple Flutter UI for reviewing completed samples on device.

How to get started

Prerequisites

• Python 3.10 or newer
• Conda or venv
• Flutter SDK
• Access to the malaria datasets used by the training scripts

Python setup

Using Conda:

conda env create -f environment.yml
conda activate malli
pip install -r requirements.txt

Using venv:

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

The pinned Python stack is intentional. The checked-in versions in requirements.txt and environment.yml are the safest place to start.

Flutter setup

flutter pub get

Train a model

python train.py --launch-dashboard

This runs the staged training pipeline defined in train.py. You can pass --config to load a JSON override file.

Evaluate saved weights

python models/evaluate.py \
  --weights models/best_mobilenetv3_small.weights.h5 \
  --data-root nih_data \
  --synthetic-root synthetic_field_ready \
  --synthetic-labels-csv labels.csv

Run the Flutter app

flutter run

Quick Python usage

from models import predict_image, CellCounter

probability, label = predict_image(
    "models/best_mobilenetv3_small.weights.h5",
    "nih_data/cell_images/Parasitized/example.png",
)

counter = CellCounter(verbose=True)
result = counter.process_image("nih_data/cell_images/Parasitized/example.png")
print(probability, label)
print(result.to_dict())

Where to get help

Start with the project docs that describe the supported workflows:

• START_HERE_TRAINING_PIPELINE.md
• DEPENDENCY_MANAGEMENT.md
• docs/CELL_COUNTER_GUIDE.md
• docs/PIPELINE_STAGES_REFERENCE.md
• docs/Roadmap.md
• docs/deep_dive/README.md
• IMPLEMENTATION_SUMMARY.md
• DETECTION_PIPELINE_IMPLEMENTATION.md

If you are debugging a specific module, the most useful starting points are data/data_loader.py, data/synthetic_data_loader.py, models/inference.py, and lib/services/image_processor.dart.

Who maintains and contributes

The project is maintained in the JFrusher/MALLI repository workspace by the MALLI project team.

Contributions should follow CONTRIBUTING.md. Keep changes focused, document behavior changes, and run the relevant Python or Flutter checks before opening a pull request.

If you are extending the ML pipeline, start from train.py and the files under models/. For mobile work, start from lib/main.dart and the files under lib/.