Explore the world of Generative Adversarial Networks (GANs) for image synthesis and translation in this 46-minute video workshop by Dr. Jan Kautz. Dive into the complexities of GANs, their applications in generating conditional facial features, and recent advancements in the field. Learn to identify and explain essential components of GANs, including Deep Convolutional versions, modify existing implementations, and design GANs for novel applications. Gain insights into recent improvements in GAN loss functions and understand the challenges of training these models. Cover topics such as image translation, conditional GANs, supervised vs. unsupervised learning, unimodal vs. multimodal approaches, and the shared latent space assumption. Ideal for data scientists, researchers, and software developers familiar with deep learning tools like Keras and TensorFlow, this workshop provides a comprehensive overview of GAN technology and its potential in image manipulation and generation.
Generative Adversarial Networks for Image Synthesis and Translation - Dr. Jan Kautz
Overview
Syllabus
Introduction
Image Translation
Training Images
Conditional Gans
Supervisor vs Unsupervised
Unimodal vs multimodal
Multiple methods
Pigs to Pigs HD
Semantic Label Map
Training Methods
Examples
Shared latent space assumption
Constraints
Weights
Results
Unsupervised Multimodal Image Translation
Conclusion
Taught by
Open Data Science
