Controlling 3D Objects in 2D Image Synthesis

• DOI: 10.1007/s42979-022-01462-w
• Publication Date: 2022-11-10T21:13:44Z
• Authors: Stephan Brehm; Florian Barthel; Rainer Lienhart
• Keywords: ddc:004; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

AbstractIn this work, we propose a method that enforces explicit control over various attributes during the image generation process in a generative adversarial net. We propose a semi-supervised learning procedure that allows us to use a quantized approximation of object orientation for learning continuous object rotations. As a result, among many other attributes, our proposed method allows us to control object orientation in scenes that are rendered according to our specifications.