Sep 9–12, 2019

Generative models for fixing image defects

Akhilesh Kumar (Adobe)
1:45pm2:25pm Wednesday, September 11, 2019
Location: 230 C

Who is this presentation for?

Machine Learning Engineer

Level

Advanced

Description

When we take any photograph using digital camera, there are several defects that are introduced in the photograph knowingly/unknowingly depending on lighting, motion etc. Some of these defects are noise defects, underexposure/overexposure or blurriness because of motion blur or camera being simply being out of focus.
It takes lot of time on image editing software to fix these defects.
We have developed a general adversarial network-based solution than can identify the region in the images that have image defects and it can fix those defects.

In the presentation I will cover following
1) What are image defects and extreme cases where it is very hard to differentiate between defective and good images?
2) Generative adversarial network and kind of GANs
3) How we locate the areas of defectiveness in the image?
4) How GANs are applied to only those areas?
5) Techniques we are developing to make GANs faster like using Bayesian network and having prior information.
6) Using the solution to fix defective videos by applying algorithm frame by frame in the videos.
7) Why GAN based approach can be better than traditional algorithms.
8) Re- using already trained classifiers as discriminator for GAN.

My presentation will mention network like Resnet 50, Inception and will give insights on how we train network.

Prerequisite knowledge

Machine learning and deep learning basics

What you'll learn

1) Cutting edge technology like GAN applied to fix photographic defects like Noise, Blur and Exposure. 2) Different variants of GAN. How GAN can be faster. 3) How deep learning-based methods are better than traditional algorithms.
Photo of Akhilesh Kumar

Akhilesh Kumar

Adobe

Akhilesh is a senior machine learning engineer at Adobe. He works in applied machine learning team at Adobe which is primarily responsible for putting deep learning models in production. Part of his job is to train, evaluate and put deep learning models in scalable systems. He is an avid reader and loves to come up with solution for wide variety of problems.

Leave a Comment or Question

Help us make this conference the best it can be for you. Have questions you'd like this speaker to address? Suggestions for issues that deserve extra attention? Feedback that you'd like to share with the speaker and other attendees?

Join the conversation here (requires login)

Contact us

confreg@oreilly.com

For conference registration information and customer service

partners@oreilly.com

For more information on community discounts and trade opportunities with O’Reilly conferences

Become a sponsor

For information on exhibiting or sponsoring a conference

Contact list

View a complete list of O'Reilly AI contacts