Computationally Efficient Approaches for Image Style Transfer

TL;DR

This paper explores various style transfer methods, focusing on reducing computational complexity through techniques like depth-wise separable convolution and interpolation, achieving similar perceptual quality with significantly faster processing times.

Contribution

The paper introduces a computationally efficient style transfer approach using innovative techniques to reduce processing time while maintaining quality.

Findings

Testing time decreased by up to 57.1%

Perceptual quality remained similar to existing methods

Multiple interpolation techniques improved efficiency

Abstract

In this work, we have investigated various style transfer approaches and (i) examined how the stylized reconstruction changes with the change of loss function and (ii) provided a computationally efficient solution for the same. We have used elegant techniques like depth-wise separable convolution in place of convolution and nearest neighbor interpolation in place of transposed convolution. Further, we have also added multiple interpolations in place of transposed convolution. The results obtained are perceptually similar in quality, while being computationally very efficient. The decrease in the computational complexity of our architecture is validated by the decrease in the testing time by 26.1%, 39.1%, and 57.1%, respectively.