Architecture data flow

This file describes the step-by-step data flow for a single inspection cycle and for retraining.

Normal inspection cycle

Sole enters camera view.
Position determination fires a trigger (sensor or AI).
Video stream → frame captures one still frame from the active camera.
Image preparing normalizes, filters, crops, resizes, and rotates the frame according to the channel config.
Main Server receives the prepared image.
Main Server routes the image to the matching YOLO instance for that channel.
YOLO instance loads weights from AI Model and runs inference.
YOLO instance returns:
- detected defect classes;
- bounding boxes;
- confidence scores.
Main Server builds an event record:
- sole ID;
- channel ID;
- timestamp;
- original image reference;
- annotated image reference;
- defect probability / class.
Main Server writes the event to DB.
WEB UI fetches and displays the event.
Expert / operator reviews the result and marks it correct or incorrect.
WEB UI sends the label back to Main Server, which updates the event in DB.

Operator selects a date range in the WEB UI retraining panel.
WEB UI requests retraining from Main Server.
Main Server queries DB for verified events in the selected range.
Main Server assembles / augments images into the Learning Dataset.
Optionally, Artificial image generator adds synthetic disturbances.
Learning module trains / fine-tunes a model on the dataset.
Learning module stores new weights in AI Model.
YOLO instances can be reloaded with the new weights (hot-reload or restart).
WEB UI shows retraining status and metrics.

Fake input data produces frames offline.
OR gate selects between real camera pipeline and fake input.
Selected frames go through Image preparing and then the same inference path.
Results are stored in DB and shown in WEB UI, exactly like production.