Wound Segmentation documentation !
Wound Segmentation is a deep learning-based tool developed to automatically detect and segment wound regions from clinical wound images. This project leverages a U-Net–style architecture with an EfficientNetB3 encoder to provide accurate binary masks for wound boundaries. It is designed to be modular, reproducible and ready for deployment.
The system supports both single image and batch mode processing, and it is equipped with configurable parameters such as prediction thresholding and flexible input formats. All core functionalities—from preprocessing to model loading, inference, and result saving—are documented and covered by automated tests, ensuring reliability and transparency.
Key features include:
Pretrained model inference using a robust U-Net backbone
Batch and single image prediction modes for flexibility in evaluation
End-to-end preprocessing and postprocessing pipeline, including aspect ratio preserving resizing, center cropping, and overlay visualization.
Command-line interface for ease of integration into larger workflows
Test coverage for all critical components with high code quality and exception handling
Fully auto-generated API documentation using Sphinx for maintainability
Lightweight—can run locally with minimal setup
Contents: