Docs

The manual and guide text from your BadBob app.

Contents

Create ML · Detection

Train an object detector from reviewed boxes and classes.

Before you start

Choose a complete published box dataset with Train and Val. Saved images with an empty annotation list are background examples and remain in their assigned split. Train must also contain annotated objects; missing revisions, invalid boxes and split overlap still block training.

Quick start

  1. Choose the dataset, model name and version. Choose Full Network or Transfer Learning.
  2. Set automatic or manual iterations and batch size. Choose an output destination, then Train.
  3. Review progress and the saved model. Evaluate it before Register model.

Functions and choices

  • Algorithm

    Full Network trains the full YOLOv2 network. Transfer Learning reuses Apple Object Print features and trains fewer parameters.

  • Estimate the budget

    Automatic iterations estimates a starting budget, not model quality.

  • Previous model

    Previous model records lineage; it does not resume a checkpoint.

  • Stop

    Stop interrupts an active run; the completion alert is optional.

  • Model Journal

    Model Journal opens the registered candidate and its provenance.

  • Use model

    Registration does not attach the project model. Use model opens a copy for inspection and confirmation before attachment.

Result

A saved Core ML detector and run receipt.

Next step

Evaluate on Val/Test and register a candidate when appropriate.

All documentation