Explore the process of making a 2D Generative Adversarial Network (GAN) 3D-aware in this 22-minute presentation by the Fellowship.ai team. Delve into the modifications required for StyleGANv2 to achieve 3D awareness, focusing on the introduction of a multiplane image style generator branch and a pose-conditioned discriminator. Learn about Generative Multiplane Images (GMPIs) and their advantages, including view consistency and dynamic adjustment of alpha maps. Discover the efficiency of this approach, which enables fast training at high resolutions. Examine experiments, evaluation metrics, and ablation studies conducted across challenging datasets like FFHQ, AFHQv2, and MetFaces. Gain insights into cutting-edge AI applications in 2D to 3D image generation, based on research by Xiaoming Zhao and colleagues.
Overview
Syllabus
Introduction
Overview
Basic Idea
Background
Alpha Branch
Experiments
Evaluation Metrics
Results
Ablation Studies
Taught by
Launchpad
Related Courses
-
Literacy Essentials: Core Concepts Generative Adversarial Network
-
Keras Deep Learning & Generative Adversarial Networks (GAN)
-
Generative Adversarial Networks (GANs)
-
Building Generative Adversarial Networks
1.0 -
Deep Learning with PyTorch : Generative Adversarial Network
-
Apply Generative Adversarial Networks (GANs)
