Is it possible to combine two different classes in a trained network?

I currently have 4 different classification classes, and I was wondering if it is possible to combine two classes into one class. For example, if I have 4 classes a, b, c, and d, if the neural network classifies an image as class a or b, the output would be 1 class, or if the network classifies it as class c or d, the output would be another class.

Answer 1

[Nihal Reddy]

I understand you want to combine two different classes to one single class when classifying an image using the nerual network.

You can do this in preprocessing step before training the network. You can combine the two classes a and b in the dataset itself and give x as the name of the new class and combine the two classes c and d in the dataset itself and give y as the name of the new class. This will help in training the neural network based on the new combined classes you have created and get the desired classification from the neural network.

Another approach would be to add a postprocessing step after the neural network classifies the image and based on the classifcation you can conditionally output the new combined class as the output.

Refer to the following for creating simple image classification network-

https://www.mathworks.com/help/deeplearning/gs/create-simple-deep-learning-classification-network.html

[Kelvin Kuang]

If I were to do the postprocessing apporoach, would the feature maps generated by the network reflect the changes, or would they still be similar to before the changes?