HomeDocumentation
Log In
Documentation

Training Detectors

The process to train a Detector is completed in two steps:

  1. Annotate the images with bounding boxes.
  2. Build the model.

If images have been uploaded with annotations (see Creating Datasets and Uploading Images), jump ahead to Building the Detector.

Bounding box annotation

After images have been uploaded into Brain Builder, they can be annotated in the Workspace. The Workspace is a feature-rich interface for drawing bounding boxes around objects of interest and labeling each one with a class.

Tagging tools

Brain Builder includes a number of tools in the Workspace to make your tagging more efficient—each of which is best suited to tag different types of objects. Tools may be selected with a mouse click or with hotkeys (noted in parentheses below).

IconTool
Bounding Box (B)
Use the bounding box tool to draw around an object within the image so you can tag the object with a class.
Brightness (R)
Use the Brightness slider to change the brightness of the image. This can help reveal details to make tagging easier. It does not change the actual image.
Contrast (C)
Use the Contrast slider to change the contrast of the image. This can help reveal details to make tagging easier. It does not change the actual image.
Undo (Ctrl+Z Windows, Cmd+Z Mac)
Redo (Ctrl+Shift+Z Windows, Cmd+Shift+Z Mac)
Use Undo and Redo to step your tagging work back and forth to quickly correct mistakes.
Select (T)
Use the Select tool to adjust anchor points, move polygons, or select polygons.
Zoom (+/-)
Use the Zoom tools to enlarge areas of the image for detailed tagging. You can zoom in, zoom out, or type a specific zoom level into the text field.

Tagging Images

  1. Use the Bounding Box tool to draw rectangles around an object in the image that you want to label with a class.
  2. After drawing the bounding box, select a Class from the right-side list to apply it to that box. If the Class has not been set up for this dataset, click the plus icon to create a new class.
  3. When you have created bounding boxes for all items in this image, click the Next > arrow in the top-right corner to move to the next image. You will be prompted to either Save & Continue or Don't Save Boxes.

Building the Detector

After the images have all been annotated, it is time to build the Detector.

  1. Navigate to the Train page using the link on the left side of the screen.
  2. On the Train screen, select the Duration of the training that you want to use.
    1. Quick training will complete the fastest but may not result in the highest accuracy.
    2. Thorough training takes the longest time to process and ensures the best results.
    3. Balanced training falls between the other two options.
      After the training duration has been selected, click the Train button.
  3. Training can take multiple hours, depending on the number of images and annotations in the dataset. If you selected the option to be emailed on completion, then Brain Builder will send an email once the training has finished.