How lmiapc/medical-fl-benchmark works
This repository benchmarks federated medical-image classification through separate traditional, personalized, elastic-aggregation, and diffusion-augmented training programs. The workflow is to prepare client-partitioned image data, train and coordinate client models, then evaluate them; the diffusion branch additionally trains a conditional generator and samples images for augmented training. Relationships below follow the excerpts, SOURCE INDEX references, or README; unsampled implementation details are not inferred.
Main entry points: dataset_define.py Dataset loading · utils.py Model utilities · fed_train_with_gen.py Augmented FL · fed_train_main.py Traditional FL
Components 14
Medical image data
- Dataset loading
data/dataset_define.py - Data partitioning
divide.py
Federated training
- Traditional FL
fed_train_main.py - Personalized FL
personalized_fed_train.py - Distillation FL
distill_fed_train.py - Elastic aggregation
elastic_aggregation.py - Augmented FL
fed_train_with_gen.py
Diffusion augmentation
- Generator training
DDPM/ddpm_train.py - Image sampling
DDPM/ddpm_inference.py - Conditional U-Net
DDPM/modules.py - Diffusion utilities
DDPM/ddpm_util.py
Shared modeling
- Model utilities
utils.py
Other components
- Benchmark user
- Medical images
Connections 21
- Benchmark user → Traditional FL: runs benchmark
README.md - Benchmark user → Personalized FL: runs benchmark
README.md - Benchmark user → Distillation FL: runs benchmark
README.md - Benchmark user → Elastic aggregation: runs benchmark
README.md - Benchmark user → Augmented FL: runs benchmark
README.md - Medical images → Dataset loading: provides images
- Traditional FL → Dataset loading: loads data
fed_train_main.py - Personalized FL → Dataset loading: loads data
personalized_fed_train.py - Elastic aggregation → Dataset loading: loads data
elastic_aggregation.py - Augmented FL → Dataset loading: loads data
fed_train_with_gen.py - Traditional FL → Model utilities: uses utilities
fed_train_main.py - Personalized FL → Model utilities: uses utilities
personalized_fed_train.py - Distillation FL → Model utilities: uses utilities
distill_fed_train.py - Elastic aggregation → Model utilities: uses utilities
elastic_aggregation.py - Augmented FL → Model utilities: uses utilities
fed_train_with_gen.py - Generator training → Diffusion utilities: loads data helpers
DDPM/ddpm_train.py - Generator training → Conditional U-Net: trains model
DDPM/ddpm_train.py - Image sampling → Conditional U-Net: samples with
DDPM/ddpm_inference.py - Image sampling → Model utilities: uses utilities
DDPM/ddpm_inference.py - Augmented FL → Image sampling: uses generated data
README.md - Data partitioning → Dataset loading: partitions data