Review: Image Style Transfer Using Convolutional Neural Networks

This article is about review of “Image Style Transfer Using Convolutional Neural Networks” (L. A. Gatys, A. S. Ecker, and M. Bethge. Image Style Transfer Using Convolutional Neural Networks. In Proc. CVPR, 2016.)

http://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf

Format for a review paper

I used specific format which is stated by Yoichi Ochiai.

1. What is the content?
2. What is the superior point comparing to previous research?
3. What is the important method?
4. How do they verify effectiveness?
5. Is there any discussion about this paper?
6. What paper should I read next?

1. What is the content?

This paper shows that VGG model is enable to render image of semantic content with style of other image.

2. What is the superior point comparing to previous research?

Compare to previous research, VGG trained model enable to separate content and style.

3. What is the important method?

CNN(Convolutional Neural Networks)’s deep layers tend to extract advanced feature. It means that deep layers can extract high-level semantic information.With this tendency, CNN model can separate content and style information.

4. How do they verify effectiveness?

To set photographs as input images for content and artworks as input images for style and to verify qualitatively with picture generated by VGG16 model.

5. Is there any discussion about this paper?

• The speed of the synthesis procedure(Taking an hour with Nvidia K40 GPU, 512pix*512pix picture)
• Subjecting some low-level noise with input images of photographs for content and style
• Difficulty to separate content and style completely

6. What paper should I read next?

• NST(L. A. Gatys, A. S. Ecker, M. Bethge, A. Hertzmann, and E. Shechtman. Controlling perceptual factors in neural style transfer. In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.)