Fast neural style
5 likes796 views
Fast Neural Style is a technique that allows neural networks to transfer artistic style from paintings to photographs in real-time. It works by using a pretrained convolutional neural network to extract feature representations from images and matching the style representations between images. The technique was improved by using instance normalization instead of batch normalization to reduce artifacts. Implementations need to address issues like border artifacts and checkerboard artifacts by using reflection padding and replacing deconvolutions with resize-convolutions. A demo shows Fast Neural Style successfully transferring styles from paintings onto photos in seconds, showing the technique's potential for neural art generation.
![Random Thoughts on Paper Implementations [KAIST 2018]](https://cdn.slidesharecdn.com/ss_thumbnails/kaist2018-180426031539-thumbnail.jpg?width=560&fit=bounds)
![[ICLR2017 Reading Meeting@DeNA] Introduction of ICLR2017](https://cdn.slidesharecdn.com/ss_thumbnails/iclr2017denaiclr2017english-170616173450-thumbnail.jpg?width=560&fit=bounds)
More Related Content
Similar to Fast neural style (20)
Recently uploaded (20)
Fast neural style
- 1. Fast Neural Style Sam Lee 2017.5.20 Tainan.py X MOUST
- 9. A Neural Algorithm of Artistic Style https://arxiv.org/abs/1508.06576 Leon A. Gatys, Alexander S. Ecker, Matthias Bethge Sep 2015
- 10. by Mark Chang Styler Transfer
- 11. Styler Transfer by Mark Chang
- 12. Image from: A Neural Algorithm of Artistic Style (Gatys et.) Layers
- 16. ICLR-2017 ύՓģ 45 ƪaʽģP 37 ƪcӖP NIPS-2016 ڕhV GAN ἰ^ 120 Σ ͬrhTᘌ Adversarial Training Mһ workshop 32 ƪ£^c GAN ֱP⣬߀ 17 ƪaʽ ģP£ 11 ƪӖP£ Arxiv Computer Science ¼s 500 ƪcWjP£ н^ 2016 Ĺ YρԴ : GANģͼ2016ȵĽչ t GAN
- 17. Generative Adversarial Networks Ian J. Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, et. Jun 2014 https://arxiv.org/abs/1406.2661
- 18. DƬԴ GAN
- 19. GAN
- 20. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks Alec Radford, Luke Metz, Soumith Chintala Jan 2016 https://arxiv.org/abs/1406.2661
- 21. DCGAN
- 22. DCGAN
- 23. DCGAN
- 24. DCGAN
- 26. GAN Fast Neural Style GAN+Style Transfer = Fast Neural Style
- 27. Perceptual Losses for Real-Time Style Transfer and Super-Resolution Johnson, Justin and Alahi, Alexandre and Fei-Fei, Li Mar 2016 https://arxiv.org/abs/1603.08155
- 28. Model
- 30. Instance Normalization: The Missing Ingredient for Fast Stylization Dmitry Ulyanov, Andrea Vedaldi, Victor Lempitsky Sep 2016 https://arxiv.org/abs/1607.08022
- 31. Instance Normalization Batch Normalization Instance Normalization
- 33. DƬԴ :https://github.com/titu1994/Fast-Neural-Style - Border Artifact ? Remove zero-padding ? Add ReflectionPadding in Input, so input & output have same size
- 34. http://distill.pub/2016/deconv-checkerboard/ - Checkerboard Artifacts ? switching deconvolutional layers for resize- convolution layers
- 36. Demo
- 37. Demo
- 38. Demo
- 39. Demo
- 40. Demo
- 41. Demo
- 42. Demo