Generative models for fixing image defects

Akhilesh Kumar (Adobe)

1:45pm–2:25pm Wednesday, September 11, 2019

Location: 230 C

Implementing AI

Secondary topics: Computer Vision, Deep Learning, Machine Learning

Who is this presentation for?

Machine learning engineers

Level

Advanced

Description

When you take any photograph using a digital camera, defects can be introduced depending on lighting, exposure, etc., which can include underexposure or overexposure, out-of-focus blur, etc. You can end up spending a lot of time on image-editing software fixing these defects.

Akhilesh Kumar investigates approaches that make it easy to identify the region of the defects. You’ll see a general adversarial network- (GAN) based solution that can automatically fix some defects, like over- or underexposure. Akhilesh details how a GAN-based solution can be better than the traditional approach of correcting images, and how this can be applied on defective images by treating videos frame by frame.

Prerequisite knowledge

An understanding of machine learning and deep learning

What you'll learn

Find out how cutting-edge technology like GAN can be used for fixing image defects
Learn about GANs
Understand how deep learning-based methods can be better than traditional algorithms

Akhilesh Kumar

Adobe

Akhilesh Kumar is a senior machine learning engineer on the applied machine learning team at Adobe, where he’s 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’s an avid reader and loves to come up with solutions for a wide variety of problems.