How to retrain a machine learning model for a multi-labels image classification with new images of already presented label or new label?

Question

Let us consider, we have several images of 3 different animals let say a Dog, a Cat and a Horse. Then we have trained our image classification model using machine learning with these 3 labels. Currently, we are using this ML model for animals image classification.

Now, whenever we have some new images of a Cat, a Horse and also of new animals Tiger and Lion, we want to add these new images into our ML model with new animals images into new labels categories and use the updated ML model for image classification. So, in current scenario, we have 5 categories of animals images.

So, how to retrain such ML model optimally? Which technologies, frameworks or platforms are best to use in this case?

Answer 1

The most promising approach would be to use a Deep neural network, especially CNN's since you're dealing with images, the idea here is to leverage Transfer Learning since this will allow you to adapt your model to new classification tasks by retraining a few layers each time you add a new label, in contrast to other methods where you have to retrain the whole model from scratch each time you add a new layer. as for the technology I would recommend using PyTorch, they have a fantastic tutorial you can start with here: tutorial