Making a 2D GAN 3D Aware

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.

Introduction
Overview
Basic Idea
Background
Alpha Branch
Experiments
Evaluation Metrics
Results
Ablation Studies