Creating custom Object Detection models (beta)

You can create IBM Watson Object Detection custom models, trained with your own images, to suit your specific application needs. This topic describes how to use the Visual Recognition model builder in IBM Watson Studio to create a custom object detection model.

Note: This feature is available as a beta and is subject to change.

Before you begin

Set up your project to use the Visual Recognition model builder.

Procedure

  1. Prepare your images
    Collect a minimum of 10 images in a .zip file, and then upload them to your project.
     
  2. Train your custom model
    In the model builder, train the model by drawing bounding boxes around objects in the images and labeling them. After you do a sufficent amout of detection and labeling, Watson can offer suggestions for labeling objects to speed up the training process.
     
  3. Test your trained model
    After your model is trained, you can use the Test area of the model builder to test images for object detection using your custom model.
     
  4. (Optional) Retrain your model
    You can improve the performance of your model by adding or removing training images, labeling more objects in the images, and then retraining the model.