In this essay we present our work for the project of the Cognitive Services course. The problem we face is the artist identification, that is the ability to recognize the author of a painting, given an image of it. Our dataset consists of about 10.000 paintings by 23 different artists. In order to resolve this task, we exploited some famous Convolutional Neural Networks (CNNs), i.e. AlexNet, GoogLeNet and ResNeXt, which come with Pytorch library. Those networks have been trained according to two distinct approaches: training them from scratch and exploiting pre-trained models using transfer learning technique. We used a virtual machine (VM) hosted by Google Cloud Platform to perform our experiments. What we have achieved is a set of results that are in line or even better with the state-of-the-art, confirming that CNNs are suitable to solve this type of task
This works has been evaluated with 28/30. Keep it mind.