Description:

Diffusion models has taken the world by storm, showcasing their prowess in generating astonishingly realistic images, leaving many intrigued by their inner workings.
They have become the backbone of modern computer vision! From Dalle 2 to Midjourney, these powerful models have revolutionized the way machines understand and process information. But what exactly are Diffusion Models, and how do they work?

In this talk, we will see theoretical foundation for how Diffusion models works and create our own training and sampling pipeline in python.

Outline of the talk

• Acquire the intuition behind Stable Diffusion Model.
• Analysis of the inner workings of Diffusion Models. Understand the math, model architecture and how it looks in the code.
• Quick overview of noise schedulers - DDPM and DDIM
• Python demo for training and sampling by generating fictional video game characters.

Prerequisites:

Basic knowledge of Python and Convolutional Neural Networks.

Video URL:

https://drive.google.com/drive/folders/1OucZSSFALyWg2wofe9B5un6yxhk887yj?usp=sharing

Speaker Info:

Mayank Khanduja is a Data Scientist at Esri R&D center with five years of experience in the industry. He has worked on the development of Generative Vision models in his current organization and has also published informative guides and blogs for the same.

Speaker Links:

https://www.linkedin.com/in/mayank-khanduja-554607130/
https://medium.com/geoai/sar-to-rgb-translation-using-cyclegan-58eb2f7af1a7
https://github.com/Esri/arcgis-python-api/blob/master/guide/14-deep-learning/how-superresolution-works.ipynb
https://github.com/Esri/arcgis-python-api/blob/master/guide/14-deep-learning/how-cyclegan-works.ipynb
https://github.com/Esri/arcgis-python-api/blob/master/samples/04_gis_analysts_data_scientists/increase-image-resolution-using-superresolution.ipynb