🧱 UNICORN Baseline Template

This repository provides a minimal working template for participating in the UNICORN Challenge.
It serves as a starting point for your own submission and implements the required boilerplate to run across all tasks in the challenge.

🔍 What It Does

This repository:

• Reads the challenge input for each task (e.g., image paths, report texts, etc.).
• Produces an output JSON file for each sample.
• Provides a dummy checksum-based feature extraction as a placeholder for real models for pathology and radiology vision tasks.

📦 This ensures your code is compatible with the challenge evaluation pipeline, even before you plug in a real model!

📁 Structure

• inference.py: Main entry point for processing inputs and generating outputs.
• vision/: Contains modules for encoding and processing pathology and radiology images.
  • pathology/: Utilities and classes for handling pathology images.
    • wsi.py: Handles whole slide images and tile extraction.
    • wsi_utils.py: Utilities for tissue detection and contour processing.
    • info.py: Displays information about multi-resolution images.
    • utils.py: Helper functions for sorting coordinates and saving features.
  • radiology/: Utilities for handling radiology images.
    • patch_extraction.py: Extracts patches from 3D radiology images.
• model/: Placeholder for model-related resources.
  • README.md: Instructions for uploading or including models.
  • a_tarball_subdirectory/: Example subdirectory for tarball resources.
• resources/: Placeholder for other resources
• requirements.in and requirements.txt: Python dependencies for the project.
• Dockerfile: Defines the container environment for running the algorithm.
• do_build.sh: Script to build the Docker container.
• do_test_run.sh: Script to test the container locally.
• do_save.sh: Script to save the container image and optional tarball for upload.

🚀 Getting Started

System requirements: Linux-based OS (e.g., Ubuntu 22.04) with Python 3.10+ and Docker installed.

Depending on your preferred way of working, you can follow our tutorial for:

1. Local development with Docker.
2. Local development with Python virtual environment [experimental].

Note: we're working on adding the steps to provide the script with the necessary inputs, stay tuned!

The local Python virtual environment does not include the steps needed to create a Docker container that can be uploaded to Grand Challenge, for that please check out the Docker tutorial.

🛠️ Customization

• Modify inference.py to implement your own feature extraction or prediction logic.
• Add your model weights to the model/ directory or upload them as a tarball to Grand Challenge.
• Update requirements.in to include additional Python dependencies and regenerate requirements.txt using pip-compile.

📄 License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.