How katherlab/swarm-learning-hpe works
This repository is a collection of medical-imaging machine-learning projects intended for experiments in HPE Swarm Learning, rather than one unified application. The main workflows train or apply models to histopathology, MRI/radiology, and image-generation tasks; the README describes sharing learned model parameters between peers while keeping raw data local. The graph preserves these distinct project branches and their documented capabilities. Several ODELIA and diffusion connections are based on SOURCE INDEX references or README descriptions; unsampled implementation details and cross-project runtime wiring are not inferred.
Main entry points: 2d_ldm.py Latent Diffusion · model/main.py MRI MIL Workflow · model/main.py ODELIA Training · ncsn_runner_cond.py Conditional Training
Components 28
Histopathology ML
- Slide Features
workspace/katherlab-gpu/model/data_utils.py - Slide Training
workspace/katherlab-gpu/model/train_function.py - CLAM Attention
workspace/katherlab-gpu/model/models/model_clam.py - GPU Experiment
workspace/katherlab-gpu/model/main.py - Legacy Pipeline
workspace/katherlab-gpu/model_old/main.py
MRI Classification
- MRI MIL Workflow
workspace/marugoto_mri/model/main.py - MIL Data
workspace/marugoto_mri/model/mil/data.py - MIL Models
workspace/marugoto_mri/model/mil/_mil.py - MIL Training Helpers
workspace/marugoto_mri/model/mil/helpers.py - Categorical Analysis
workspace/marugoto_mri/model/categorical.py - ROC Evaluation
workspace/marugoto_mri/model/roc.py
ODELIA MRI
- ODELIA Training
workspace/odelia-breast-mri/model/main.py - Model Configuration
workspace/odelia-breast-mri/model/env_config.py - 3D Model Selection
workspace/odelia-breast-mri/model/model_selector.py - MRI Data Pipeline
workspace/odelia-breast-mri/model/data/datamodules/datamodule.py - MRI Prediction
workspace/odelia-breast-mri/model/predict.py
Score Generation
- Score Model Runner
workspace/score-based-model/model/main.py - Conditional Training
workspace/score-based-model/model/runners/ncsn_runner_cond.py - NCSN Network
workspace/score-based-model/model/models/ncsnv3.py - Image Datasets
workspace/score-based-model/model/datasets/__init__.py - Denoising Loss
workspace/score-based-model/model/losses/dsm.py
Latent Diffusion
- Latent Diffusion
workspace/swag-latent-diffusion/model/2d_ldm.py - Medical Image Datasets
workspace/swag-latent-diffusion/model/dataset/fastmri_brain.py - Diffusion Inference
workspace/swag-latent-diffusion/model/generative/inferers/inferer.py - Perceptual Loss
workspace/swag-latent-diffusion/model/generative/losses/perceptual.py - Adversarial Loss
workspace/swag-latent-diffusion/model/generative/losses/adversarial_loss.py - Diffusion U-Net
workspace/swag-latent-diffusion/model/generative/networks/nets/diffusion_model_unet.py
Other components
- Researcher
Connections 29
- Researcher → GPU Experiment: configures experiment
workspace/katherlab-gpu/model/main.py - GPU Experiment → Slide Features: uses features
workspace/katherlab-gpu/model/main.py - GPU Experiment → Slide Training: uses training
workspace/katherlab-gpu/model/main.py - Researcher → Legacy Pipeline: starts experiment
- Legacy Pipeline → Slide Features: loads data
workspace/katherlab-gpu/model_old/main.py - Researcher → MRI MIL Workflow: starts workflow
workspace/marugoto_mri/model/main.py - MRI MIL Workflow → MIL Training Helpers: runs training
workspace/marugoto_mri/model/main.py - MRI MIL Workflow → Categorical Analysis: aggregates predictions
workspace/marugoto_mri/model/main.py - MRI MIL Workflow → ROC Evaluation: plots ROC
workspace/marugoto_mri/model/main.py - MIL Training Helpers → MIL Data: uses data
workspace/marugoto_mri/model/mil/helpers.py - MIL Training Helpers → MIL Models: uses MIL
workspace/marugoto_mri/model/mil/helpers.py - MIL Models → MIL Data: builds datasets
workspace/marugoto_mri/model/mil/_mil.py - Researcher → ODELIA Training: starts training
workspace/odelia-breast-mri/model/main.py - ODELIA Training → MRI Data Pipeline: uses datamodule
workspace/odelia-breast-mri/model/main.py - ODELIA Training → Model Configuration: uses configuration
workspace/odelia-breast-mri/model/main.py - Model Configuration → MRI Prediction: references prediction
workspace/odelia-breast-mri/model/env_config.py - Model Configuration → MRI Data Pipeline: references datasets
workspace/odelia-breast-mri/model/env_config.py - ODELIA Training → 3D Model Selection: selects model
workspace/odelia-breast-mri/model/main.py - Researcher → Score Model Runner: configures run
workspace/score-based-model/model/main.py - Score Model Runner → Conditional Training: dispatches runner
workspace/score-based-model/model/main.py - Conditional Training → NCSN Network: trains network
workspace/score-based-model/model/runners/ncsn_runner_cond.py - Conditional Training → Image Datasets: loads datasets
workspace/score-based-model/model/runners/ncsn_runner_cond.py - Conditional Training → Denoising Loss: computes loss
workspace/score-based-model/model/runners/ncsn_runner_cond.py - Researcher → Latent Diffusion: configures training
- Latent Diffusion → Medical Image Datasets: loads dataset
workspace/swag-latent-diffusion/model/2d_ldm.py - Latent Diffusion → Diffusion Inference: uses inferer
workspace/swag-latent-diffusion/model/2d_ldm.py - Latent Diffusion → Perceptual Loss: uses loss
workspace/swag-latent-diffusion/model/2d_ldm.py - Latent Diffusion → Adversarial Loss: uses loss
workspace/swag-latent-diffusion/model/2d_ldm.py - Latent Diffusion → Diffusion U-Net: uses network
workspace/swag-latent-diffusion/model/2d_ldm.py