High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs

Explore high-resolution image synthesis and semantic manipulation using conditional GANs in this 34-minute lecture from the University of Central Florida. Delve into the pix2pix baseline and learn techniques for improving photorealism and resolution through coarse-to-fine generation and multi-scale discriminators. Discover the use of instance maps and the process of learning instance-level feature embeddings. Examine quantitative comparisons, perceptual studies, and human evaluations of generated images. Compare generator and discriminator performances, and gain insights into interactive object editing capabilities.

Intro
Motivation
pix2pix Baseline
Improving Photorealism and Resolution: Coarse to Fine Genera
Improving Photorealism and Resolution: Multi-Scale Discriminators
Using Instance Maps
Learning an Instance Level Feature Embedding
Results: Quantitative Comparison
Results: Perceptual Study
Results: Human Perceptual Study - Unlimited Tin
Results: Human Perceptual Study - Limited Time
Results: Generator Comparison
Results: Discriminator Comparison
Interactive Object Editing