Retraining a Visual Recognition custom model

There are many reasons why you might want to retrain your IBM Watson Visual Recognition custom model, such as: testing of a new model indicates performance is not as good as you need; or current input data for an older model has drifted away from the data used to train the model. In IBM Watson Studio, you can “retrain” a model by using the Visual Recognition model builder, which has the effect of deleting the original model and creating a new model with a new model ID.

Before you begin

 

Procedure

  1. From the Assets page of your project, open the Visual Recognition model builder by clicking on your custom model.
     
  2. Click Edit and Retrain.
     
  3. Edit your model: remove training images from your model, add more training images to your model, or change the class with which a training image is associated.
     
  4. Click Train Model.
     
  5. Wait for the retraining process to finish. You can leave the Visual Recognition model builder while the process runs. You can monitor the status of the training process by looking at the status of the model on the Assets page of your project.

 

Updating your apps

When you retrain a Visual Recognition model by using the Visual Recognition model builder in Watson Studio, the original model is deleted and a new model is created with a new model ID. This means that you must update the model ID in any apps that use the model to classify images.

 

Watch this short video to see how to retrain a model to add images to the negative class and reclassify images.

Figure 1. Video iconRetrain a model to add images and reclassify images
This video shows you how to retrain a model to add images and reclassify images.