This repository contains the code, dataset utilities, model definitions, and evaluation scripts used in the paper.
config/- Central configuration files for datasets, cache paths, and model-specific settings.data/- Data loading, caching, and augmentation helpers.dataset/- PyTorch dataset implementations for cast, paired, latent, and similarity-based data.downstream_tasks/- Scripts related to the downstream task evaluation.extreme_error_predictions/- Utilities for measuring extreme prediction errors.media_readme/- Figures and other assets used in the README.metadata/- Metadata helpers and supporting files.model/- Core model implementations, blocks, generators, discriminators, diffusion modules, and classifiers.radiologists_test/- Materials for the radiologist evaluation study.results/- Experiment results.testers/- Evaluation and testing scripts for trained models.trainer/- Training loops and trainer classes for classifiers and image models.utils/- Shared utilities for network setup, preprocessing, schedulers, evaluation, and file handling.
train.py- Main training entry point for image-to-image models.train_classifier.py- Training script for the classifier.train_sanitized.py- Training pipeline for sanitized data.infer_classifier.py- Classifier inference script.infer_generated_images_region.py- Classifier inference for the extreme error metric calculation.
