Chipemba/maize-leaf-disease-cv

Efficiency focused robustness evaluation for maize leaf disease classification using Southern African field images. # Lightweight Computer Vision for Maize Leaf Disease Detection ## Project Summary This project evaluates maize leaf disease classifiers for Southern African field conditions. The contribution is an efficiency-aware robustness analysis across classification accuracy, computational efficiency and robustness to field-image variation. ## Research Question Can lightweight computer vision models classify maize leaf diseases from Southern African field images with acceptable accuracy while offering better computational efficiency and robustness than larger transfer learning models? ## Project Structure ```text maize-leaf-disease-cv/ ├── .github/ │ └── workflows/ # GitHub Actions CI ├── configs/ # Model training configurations ├── data/ │ ├── raw/ # Local raw dataset (not committed) │ └── splits/ # Train/validation/test CSV files ├── demo/ │ ├── app.py # Streamlit ONNX inference demo │ └── sample_images/ # Demo images ├── notebooks/ # Exploratory notebooks ├── models/ # Trained PyTorch checkpoints (local if ignored) ├── onnx/ │ └── mobilenet_v2.onnx # Exported deployment model ├── presentation/ │ └── final_demo_screenshots/ # Final presentation/demo evidence ├── reports/ # Project reports and documentation ├── results/ # Evaluation and robustness outputs ├── src/ │ ├── data/ # Validation, EDA and dataset splitting │ ├── models/ # Model definitions/factory │ ├── training/ # Training scripts │ ├── evaluation/ # Evaluation, efficiency and robustness │ └── deployment/ # ONNX export utilities ├── tests/ # Automated tests ├── README.md ├── requirements.txt └── LICENSE ## Dataset ### Local Dataset Layout Raw image data is not committed to GitHub. After downloading the dataset, place files …