How andersonmaina/facs works
FACS is a Flask-based ultrasound image anomaly-analysis service, with separate scripts and artifacts for preparing normal-image features and training category-specific autoencoders. The main user flow accepts an annotated image from a PC or Telegram client, preprocesses it, runs inference, and returns an anomaly score and diagnosis. Source excerpts establish the API-to-model call and model selection/inference; some crop and response details are visible only in partial excerpts, and no README was provided.
Main entry points: app.py Annotation API · model.py Prediction Pipeline · feature_extractor.py Feature Extraction · train_autoencoder.py Autoencoder Training
Components 14
API and Results
- Annotation API
app.py - Image Decode and Crop
app.py - Score and Diagnosis
app.py
Anomaly Inference
- Prediction Pipeline
model.py - View and Category Models
model.py - ResNet Features
model.py - Reconstruction and Error
model.py - Trained Weights
models
Feature Preparation
- Feature Extraction
feature_extractor.py - Normal Feature Arrays
features
Model Training
- Autoencoder Training
train_autoencoder.py
Other components
- PC Client
- Telegram Client
- Annotated Images
Connections 14
- PC Client → Annotation API: submits annotation
app.py - Telegram Client → Annotation API: uploads image
app.py - Annotation API → Image Decode and Crop: decodes and crops
app.py - Annotation API → Prediction Pipeline: calls predict
app.py - Prediction Pipeline → View and Category Models: selects weights
model.py - View and Category Models → Trained Weights: loads state
model.py - Prediction Pipeline → ResNet Features: extracts features
model.py - Prediction Pipeline → Reconstruction and Error: reconstructs features
model.py - Reconstruction and Error → Prediction Pipeline: returns error
model.py - Annotation API → Score and Diagnosis: builds response
app.py - Feature Extraction → Annotated Images: reads images
feature_extractor.py - Feature Extraction → Normal Feature Arrays: saves features
feature_extractor.py - Autoencoder Training → Normal Feature Arrays: loads features
train_autoencoder.py - Autoencoder Training → Trained Weights: saves weights
train_autoencoder.py