LRNavin/painters_by_numbers

In this project we have evaluated approaches to the challenge proposed on Kaggle named {Painter by Numbers}. The objective of this challenge is to distinguish whether two paintings were created by the same artist in the process of pairwise comparison. In a broad sense this could improve the identification of forgeries based on learned {artist style}. The dataset for the challenge is a collection of paintings from WikiArt.org {http://wikiart.org}. After analyzing approaches taken by other competitors, we have identified a gap that could be explored: creating a network that not only learns the style from artists included in the provided dataset, but also is able to give a correct verdict for an artist that are not included in the training data. This creates an additional layer of complexity as the network has to facilitate ability to {generalize to unseen artists}. To tackle the problem at hand, a neural network architecture called the siamese network was used.