Hi Everyone I am a beginner in deep learning and doing a project on deep learning for my college. I want to train a CNN that can classify three classes of Skin Cancer namely Melanoma, Sebborhic Keratosis and Nevus. I have taken dataset from ISIC Challenge 2017. The data only contained 2750 images and was highly biased. Biased because out of 2750 images 1320 images were belonging to single class. Anyways I have balanced the classes by using some image transformations for uplifting the classes with less images and augmented the data. At the end I got 10000 images for training and 400 images for testing. I want to train a model from a scratch as standard models are not giving good accuracies.
Is it feasible to train a CNN model from scratch and get good accuracies with 10000 images?? If yes then how to approach for building such model .