COV877: Special Module on Visual Computing

Generative AI for Visual Content Creation: Image, Video, and 3D

IIT-Delhi

Course Information

The generation of visual content is a fundamental aspect of computer graphics that has been the focus of extensive research for decades. Recent advancements in neural representations, generative modelling and differentiable rendering techniques have propelled this field of content generation forward. As a result, computational systems that can convert simple user inputs like texts into image, text/images into videos/3D shapes have gained significant attention, empowering non-expert users to easily create visual content directly from such inputs. In this course, we will cover the fundamentals of neural representations for images, video and 3D content and generative modelling for image, video and 3D generation under different training paradigms e.g., paired 3D data, unpaired 3D data, without 3D data. .


Course Schedule

This is a tentative schedule for the course. Please check for updates regularly.

(9 Lectures, 1.5 hrs each).


  • Lecture 1: Genarative Adversarial Network (GAN) [PDF]
  • Lecture 2: StyleGAN and Normalizing Flows [PDF]
  • Lecture 3: Variational Autoencoder (VAE & VQVAE) [PDF]
  • Lecture 4: Score Based Diffusion Model [PDF]
  • Lecture 5: Probability Flow ODE and Hands-on Score-Based Diffusion Model
  • Lecture 6: Denoising Diffusion Probabilistic Models, and Conditional Generation [PDF]
  • Lecture 7: Image Generation and Editing (GLIDE, Dall-E, Stable Diffusion, SDEdit) [PDF]
  • Lecture 8: 3D Generation
  • Lecture 9: Video Generation
Assignment and Exam
  • Assignment Score-Based Diffusion Model [PDF], Deadline 23rd April, 11:59PM
Resources
  1. Specific references (papers, blogs etc) are added to each slides, look at the footnote of each slides.
  2. Understanding Deep Learning by Simon J D Prince Chapter 15 (GAN), 16(Normalizing Flows) and 17(VAEs) [https://udlbook.github.io/udlbook/]
Contact Information
  • Instructor: Dr. Lokender Tiwari - lokender.work@gmail.com