For the face model to work correctly, negative images are also required, not to be used to create a class (we will cover this in an upcoming section) within the created classifier, but to define what the updated classifier is not. Negative example files should not contain images that have the subject of any of the positive classes (happy and sad). In essence, the face images in this group should be perhaps considered to be neutral. You only need to specify one negative example file.
Because you want to give the model examples of what not to look for, you must provide the negative class. Providing a ML model with all positive images would mean that it would just assume that everything is positive and produce a risky result.
So finally, our initial negative model training data is shown in the following screenshot:
