Artistic Neural Style Transfer Algorithms with Activation Smoothing

TL;DR

This paper enhances neural style transfer by re-implementing various NST methods and introducing activation smoothing with ResNet, significantly improving stylization quality through extensive experiments.

Contribution

It introduces activation smoothing with ResNet into neural style transfer, improving stylization quality and providing comprehensive re-implementations of NST methods.

Findings

Activation smoothing greatly improves stylization quality

ResNet with smoothing outperforms traditional CNNs in NST

Experimental results validate the effectiveness of the proposed method

Abstract

The works of Gatys et al. demonstrated the capability of Convolutional Neural Networks (CNNs) in creating artistic style images. This process of transferring content images in different styles is called Neural Style Transfer (NST). In this paper, we re-implement image-based NST, fast NST, and arbitrary NST. We also explore to utilize ResNet with activation smoothing in NST. Extensive experimental results demonstrate that smoothing transformation can greatly improve the quality of stylization results.