How dzp2095/fcu works
Project type and purpose
• This is a research‐focused federated learning (FL) framework specialized for medical imaging, specifically enabling “client unlearning” in FL across two datasets: ICH (intracranial hemorrhage) and ISIC (skin lesions).
• Purpose: preprocess raw medical images, partition them into FL clients, run federated training with unlearning capability, and evaluate results.
File structure
• Top‐level:
• configs/ : YAML configs for different dataset runs and scripts
• scripts/ : dataset‐specific FL preprocessing pipelines
• src/ : core FL code (datasets, fl, model, modules, tasks, evaluation, utils)
• train.py : CLI entry point for launching FL training
• requirements.txt, .gitignore
• src breakdown:
• src/datasets : dataset wrappers & sampling
• src/fl : FL server & client implementations
• src/model : network architectures (DenseNet, generic net)
• src/modules : training loops, hooks, defaults
• src/tasks : task factories/registry for different dataset tasks
• src/evaluation : binary/multiclass evaluation strategies
• src/utils : argument parsing, device selection, logging
README
• Describes data download steps (Kaggle links), preprocessing, FL dataset generation, and training commands.
• Points to two main workflows (ICH and ISIC) driven by separate config files.
• Mentions paper citation.