How gregorator/medical-image-enhancement-system works
This is a CT-image denoising product spanning NetCDF dataset preparation, Noise2Noise training, ONNX deployment, and an interactive Streamlit studio backed by a FastAPI service. The primary user workflow is to inspect or batch-submit images in the UI, which calls the API for denoising; users can also switch inference configuration and request heatmaps or unload the model. Training and dataset preparation are separate workflows. Some links rely on README descriptions or the source index because those implementation excerpts were not provided.
Main entry points: train.py Training entry · onnx_inference.py Inference manager · api_client.py API client · factories.py Training factories
Components 26
Streamlit Studio
- Dashboard
app/pages/0_Dashboard.py - Image Inspector
app/pages/1_Image_Inspector.py - Batch Processor
app/pages/2_Batch_Processor.py - Page routing
app/routes.py - Model controls
app/components/sidebar.py - API client
app/utils/api_client.py
Inference Service
- REST API
api/api_server.py
Dataset Pipeline
- Dataset preparation
src/datasets/make_dataset.py - NetCDF conversion
src/datasets/preparation.py - Image datasets
src/datasets/dataset.py
Training
- Training entry
scripts/train.py - Run configuration
configs/config.py - Training factories
src/utils/factories.py - Noise transforms
src/transforms/noise.py - Noise2Noise U-Net
src/models/noise2noise.py - N2N trainer
src/trainers/n2n.py - Training engine
src/trainers/engine.py - Evaluation metrics
src/evaluation/metrics.py - Experiment logger
src/utils/logger.py
ONNX Deployment
- Inference manager
onnx/onnx_inference.py - GPU tiling
onnx/inference_gpu.py - CPU inference
onnx/inference_cpu.py - ONNX wrapping
onnx/onnx_wrappers.py - ONNX export
onnx/export_onnx.py - Bundled ONNX models
onnx/models
Other components
- Image user
Connections 26
- Image user → Dashboard: opens studio
README.md - Image user → Image Inspector: inspects images
README.md - Image user → Batch Processor: submits batches
README.md - Dashboard → Page routing: registered by
app/pages/0_Dashboard.py - Image Inspector → API client: requests inference
app/pages/1_Image_Inspector.py - Batch Processor → API client: requests inference
app/pages/2_Batch_Processor.py - Model controls → API client: updates settings
app/components/sidebar.py - Model controls → Page routing: uses pages
app/components/sidebar.py - API client → REST API: calls REST endpoints
README.md - REST API → Inference manager: dispatches inference
api/api_server.py - Inference manager → GPU tiling: selects CUDA backend
onnx/onnx_inference.py - Inference manager → CPU inference: selects CPU backend
onnx/onnx_inference.py - GPU tiling → ONNX wrapping: uses wrapper
onnx/inference_gpu.py - Dataset preparation → NetCDF conversion: converts blocks
src/datasets/make_dataset.py - Training entry → Run configuration: loads settings
scripts/train.py - Training entry → Image datasets: loads samples
scripts/train.py - Training entry → Training factories: builds components
scripts/train.py - Training factories → Noise transforms: selects transforms
src/utils/factories.py - Training factories → Noise2Noise U-Net: builds models
src/utils/factories.py - Training entry → N2N trainer: starts training
scripts/train.py - N2N trainer → Training engine: extends engine
src/trainers/n2n.py - N2N trainer → Evaluation metrics: computes evaluation
src/trainers/n2n.py - Training engine → Experiment logger: logs progress
src/trainers/engine.py - ONNX export → Noise2Noise U-Net: exports network
onnx/export_onnx.py - ONNX export → Bundled ONNX models: writes model
README.md - Inference manager → Bundled ONNX models: loads model
README.md