When Cartoon Meets Anime Distinguishing Animation Styles with Convolutional Neural Networks

There are various myths surrounding the pop culture, though each of them can be quite fascinating at times, one of them in particular, “Are Anime Cartoon?” succeeds to keep our interest while compelling us to venture deep into a typical machine learning algorithm, Convolutional Neural Network (CNN) to unravel the same. We thus seek to provide concrete proofs against the false speculations that claim anime and cartoons are all the same. Moreover, we would like to see if a machine can learn to classify two different types of animation formats. And to top it all, since to the best of our knowledge, this peculiar problem remains almost unaddressed, it becomes more than compelling for us to spend our resources on this endeavor.After much deliberation, we decided to choose Convolutional Neural Networks to base our research on. Moreover, as evident from various researches in the field of image classification, unlike traditional methods like multilayer perceptron, Convolutional Neural Networks extracts successively larger features in a hierarchical set of layers and thus, is best suited for our problem consisting of images with a large number of variations. For our dataset, we collected over 25,000 images from various Cartoon and Anime shows. For training, we chose 12,000 images from both categories, and the remaining images which constitute a mixture of Anime and Cartoon are used to test our model. After running multiple tests, we are quite satisfied with our model, as it has training accuracy of over 97.4% and validation accuracy of over 97.3%. This allows us to prove our initial hypothesis that anime and cartoons are, indeed, different. Moreover, since our problem involves classifying a large number of varying images, we also need to take into consideration computational complexity and discuss how various techniques can be used to optimize our model without increasing the need for computing resources.