Hi!
I’m trying to train my own image classifier for Lesson 2 assignment and I was wondering if the images from the training set should represent single objects. For example, suppose I want to train a model to classify types of video games. On Google Images it is not uncommon to fnd pictures such as
which represents two different video games (or classes). Is it OK to use such images to train? Or the model might just get “confused” resulting in high loss?
The same goes for drawings. Should I mix drawings of my classes with the real thing (e.g. drawings of video games and photos of video games on the training set)?