Hello! I am new at Machine Learning and Vision, and since WWDC I have been interested in the image classification Apple showed off as well as CreateML. I immediately created a very simple ML model that classified between a hamburger, fries, or a cup of soda, and worked well. I never looked back into it, as I could not think of a use for it. Recently, I have been playing again with it, but I have not had much success with classifying objects.
My first attempt was classifying a traffic light to determine if it is red, yellow, or green, but it was not able to successfully classify images of other traffic lights. My second attempt was classifying 3 hand gestures: holding 1 finger up, 2 fingers up, and 3 fingers up. I had pictures of the hand from 5 different angles (top, front, back, right, left), and again, it was not able to classify them correctly.
After this, I kept wondering if CoreML classifies an image by color, shape, size, or all and if there is a way to only classify by only one, such as color in the traffic light example, or shape as in the hand gesture example. I was using CreateML as specified in this documentation: https://developer.apple.com/documentation/vision/training_a_create_ml_model_to_classify_flowers
Any help or suggestions on how that could be achieved would be highly appreciated. Thanks 🙂