Transfer Learning for Illustration Classification

TL;DR

This paper introduces a transfer learning approach using a modified VGG19 network to improve illustration image classification, achieving high accuracy with less training data and maintaining object recognition in photos.

Contribution

The paper presents two novel models that adapt pre-trained deep networks for illustration classification, leveraging transfer learning to enhance performance and reduce data requirements.

Findings

Achieved 86.61% top-1 accuracy on illustration dataset.

Achieved 97.21% top-5 accuracy on illustration dataset.

Model retains ability to recognize objects in photographs.

Abstract

The field of image classification has shown an outstanding success thanks to the development of deep learning techniques. Despite the great performance obtained, most of the work has focused on natural images ignoring other domains like artistic depictions. In this paper, we use transfer learning techniques to propose a new classification network with better performance in illustration images. Starting from the deep convolutional network VGG19, pre-trained with natural images, we propose two novel models which learn object representations in the new domain. Our optimized network will learn new low-level features of the images (colours, edges, textures) while keeping the knowledge of the objects and shapes that it already learned from the ImageNet dataset. Thus, requiring much less data for the training. We propose a novel dataset of illustration images labelled by content where our optimized architecture achieves $\textbf{86.61\%}$ of top-1 and $\textbf{97.21\%}$ of top-5 precision. We additionally demonstrate that our model is still able to recognize objects in photographs.