Hi, I am trying to fine tune a CNN (Googlenet) and train a new dataset (photos of various landscapes) in order to classify new images. I was wondering if there was a way to select specific images for the training dataset, instead of randomly splitting the whole dataset into training/testing images ? For example, with three classes : select images 1-10 of class 1, images 1-10 of class 2 and images 1-10 of class 3 for the training dataset...and the rest for testing.
Does anyone know of a pratical example (which I may have overlooked) that uses this approach ? I would really appreciate it.
Thank you !