2013 Poster Sessions : Example-based Synthesis of 3D Object Arrangements

Student Name : Matthew Fisher
Advisor : Patrick Hanrahan
Research Areas: Graphics/HCI
The creation of large, 3D environments is a significant bottleneck for film and video game production. To make progress towards reducing this bottleneck, we present a method for synthesizing 3D object arrangements from examples. Our algorithm can synthesize a diverse set of plausible new scenes given only a few examples and requires no additional inputs from the user. These capabilities are enabled by three novel contributions. First, we introduce a probabilistic model for scenes based on Bayesian networks and Gaussian mixtures that can be trained from a small number of input examples. Second, we develop a clustering algorithm that groups objects occurring in a database of scenes according to their local scene neighborhoods. These contextual categories allow the synthesis process to treat a wider variety of objects as interchangeable. Third, we train our probabilistic model on a mix of user-provided examples and relevant scenes retrieved from the database. This mixed model learning process can be controlled to introduce additional variety into the synthesized scenes. We evaluate our algorithm through qualitative results and a perceptual study in which participants judged synthesized scenes to be highly plausible, as compared to hand-created scenes.

Matthew Fisher is a PhD candidate in Computer Science at Stanford University. His research in computer graphics focuses on combining large databases with machine learning approaches to support complex creative tasks. Previously he has worked on the Direct3D graphics kernel at Microsoft and his PhD at Stanford was supported with a fellowship from the Hertz Foundation. He earned his Masters in Computer Science at Stanford University, and a Bachelors degree in Computer Science at the California Institute of Technology.