Style Transfer with PyTorch

This course covers the important aspects of neural style transfer, a technique for transforming images, and discusses Generative Adversarial Networks in order to efficiently create realistic images and videos.
Course info
Level
Advanced
Updated
Aug 2, 2019
Duration
1h 49m
Table of contents
Description
Course info
Level
Advanced
Updated
Aug 2, 2019
Duration
1h 49m
Your 10-day individual free trial includes:

Expert-led courses

Keep up with the pace of change with thousands of expert-led, in-depth courses.
Description

Style transfer refers to the use of a neural network to transform an image so that it comes to artistically resemble another image, while still retaining its original content. Neural style transfer is fast becoming popular as a way to change the aesthetics of an image. In this course, Style Transfer with PyTorch, you will gain the ability to use pre-trained convolutional neural networks (CNNs) that come out-of-the-box in PyTorch for style transfer. First, you will learn how style transfer involves a style image as well as a content image, and a pretrained neural network that usually does not change at all during the training process. Next, you will discover how intermediate layers of the CNN are designated as style layers of interest and content layers of interest. Then, you will explore the minimization of two loss functions - a style loss and a content loss. Finally, you will delve into leveraging a new and much-hyped family of ML models, known as Generative Adversarial Networks (GANs) to create realistic images and videos. When you’re finished with this course, you will have the skills and knowledge to perform neural style transfer to get images that combine content and artistic style from two different inputs and use GANs to generate realistic images from noise.

About the author
About the author

A problem solver at heart, Janani has a Masters degree from Stanford and worked for 7+ years at Google. She was one of the original engineers on Google Docs and holds 4 patents for its real-time collaborative editing framework.

More from the author
Summarizing Data and Deducing Probabilities
Intermediate
2h 50m
Jul 8, 2021
More courses by Janani Ravi
Section Introduction Transcripts
Section Introduction Transcripts

Course Overview
Hi. My name is Janani Ravi, and welcome to this course on Style Transfer with PyTorch. A little about myself. I have a masters degree in electrical engineering from Stanford and have worked at companies such as Microsoft, Google, and Flipkart. At Google, I was one of the first engineers working on real-time collaborative editing in Google Docs, and I hold four patents for its underlying technologies. I currently work on my own startup, Loonycorn, a studio for high-quality video content. Style Transfer refers to the use of a neural network to transform an image so that it comes to artistically resemble another image while still retaining its original content. In this course, you will gain the ability to use pretrained convolutional neural networks that come out of the box in PyTorch for style transfer. First, you will learn how style transfer involves a style image, as well as a content image and a pretrained neural network that unusually does not change at all during the training process. Next, you will discover how intermediate layers of the CNN are designated as style layers of interest and content layers of interest. Then two loss functions are minimized, a style loss and a content loss. The minimization of these two losses leads to an output target image with content from the content image and style from the style image. Finally, you'll explore how to leverage a new and much hyped family of ML models known as GANs, or generative adversarial networks. You'll see how GANs can be used to create realistic images and videos of virtually anything. When you're finished with this course, you will have the skills and knowledge to perform neural style transfer to get images that combine content and artistic style from two different inputs and use GANs to generate realistic images from noise.