Docs

The manual and guide text from your BadBob app.

Contents

Evaluation · Create ML

Inspect performance on held-out reference images and find concrete mistakes.

Before you start

A completed Create ML result and available evaluation images.

Quick start

  1. Open Evaluation and choose Create ML in the backend selector.
  2. Select the run and evaluation data, then run evaluation. Inspect missing objects, extra objects and wrong classes.
  3. Inspect individual predictions and preserve useful failure evidence for review.

Functions and choices

  • Model

    Select a model by name and version; the short run ID is shown for identification and is used when a name is unavailable.

  • Choose the split

    Select a dataset and split before Evaluate. Train is inspection only; Val supports tuning and Test is held out. Use consistent splits and thresholds when comparing results. Training loss alone does not establish detection or mask quality.

  • Browse and select

    Click a list row, then use Up/Down to browse the filtered images. The separate switches choose images for Failure Bank; browsing does not change that selection. Changing the filter keeps the active image visible, or selects the first matching image.

  • Compare annotations

    Human annotations (solid lines) and Model suggestions (dashed lines) use project class colors and can be shown independently to inspect overlapping outlines.

  • Handle errors

    Load and action errors stay above the image workspace until dismissed with the close button or a different project is opened. A successful background refresh does not hide the message.

Result

Saved evaluation evidence plus examples that explain the metrics.

Next step

Compare against an earlier model or review mistakes in Failure Bank.

All documentation