How alvin20082003/quantum-retina-x works
QuantumRetina-X is a retinal diabetic-retinopathy grading system with a web interface, an inference API, and a research workflow for training and evaluating ViT and hybrid ViT–quantum models. The user flow is image upload, preprocessing, model inference, and a result containing severity, risk and recommendation information, with Grad-CAM and TCAV-derived explanations. The graph also preserves dataset preparation, model variants, experiment metrics and plotting. The README is empty; unsampled links are omitted where the source index alone does not establish a call.
Main entry points: main.py Prediction API · hybrid.py Hybrid ViT–VQC · train.py Model training · lowdata_experiment.py Low-data study
Components 20
Clinical interface
- Upload and results UI
frontend/src/App.tsx
Inference and explanations
- Prediction API
backend/main.py - Image transforms
data/transforms.py - Grad-CAM overlays
backend/main.py - TCAV and DME results
backend/main.py
Model architecture
- ViT classifier
models/backbone.py - Hybrid ViT–VQC
models/hybrid.py - Quantum layer
models/quantum.py - VQC-only variant
models/vqc_only.py
Training and evaluation
- APTOS data loaders
data/loaders.py - Model training
training/train.py - Low-data study
training/lowdata_experiment.py - Cross-dataset studies
training/cross_dataset.py - Model evaluation
eval_hybrid.py - Evaluation metrics
training/metrics.py - Experiment plots
generate_all_plots.py
Shared foundations
- Configuration
utils/config.py - Reproducible seeding
utils/seed.py - Experiment logger
utils/logger.py
Other components
- Clinician
Connections 29
- Clinician → Upload and results UI: uploads image
frontend/src/App.tsx - Upload and results UI → Prediction API: requests prediction
- Prediction API → Image transforms: preprocesses image
backend/main.py - Prediction API → ViT classifier: runs baseline
backend/main.py - Prediction API → Hybrid ViT–VQC: runs hybrid
backend/main.py - Prediction API → Grad-CAM overlays: generates overlay
backend/main.py - Prediction API → TCAV and DME results: derives result details
backend/main.py - Hybrid ViT–VQC → Quantum layer: calls quantum layer
models/hybrid.py - VQC-only variant → Quantum layer: calls quantum layer
models/vqc_only.py - APTOS data loaders → Image transforms: builds transforms
data/loaders.py - Model training → APTOS data loaders: loads batches
training/train.py - Model training → ViT classifier: trains baseline
training/train.py - Model training → Evaluation metrics: computes scores
training/train.py - Low-data study → Image transforms: builds transforms
training/lowdata_experiment.py - Low-data study → ViT classifier: trains baseline
training/lowdata_experiment.py - Low-data study → Hybrid ViT–VQC: trains hybrid
training/lowdata_experiment.py - Low-data study → Evaluation metrics: computes scores
training/lowdata_experiment.py - Low-data study → Experiment logger: logs results
training/lowdata_experiment.py - Cross-dataset studies → Hybrid ViT–VQC: evaluates hybrid
training/cross_dataset.py - Cross-dataset studies → Evaluation metrics: computes scores
training/cross_dataset.py - Model evaluation → APTOS data loaders: loads evaluation data
eval_hybrid.py - Model evaluation → Hybrid ViT–VQC: evaluates model
eval_hybrid.py - Model evaluation → Evaluation metrics: computes scores
eval_hybrid.py - Model evaluation → Experiment logger: logs results
eval_hybrid.py - Model training → Configuration: loads settings
training/train.py - Model training → Reproducible seeding: sets seed
training/train.py - Experiment plots → APTOS data loaders: uses dataset loaders
generate_all_plots.py - Experiment plots → Hybrid ViT–VQC: plots hybrid results
generate_all_plots.py - Experiment plots → Configuration: loads settings
generate_all_plots.py