Learning Geometry

Summary: We introduce a deep learning approach for grouping discrete graphical patterns. Our approach learns a grouping measure defined over pairs of pattern elements using a convolutional neural network architecture. Our network encodes element shape, context, symmetries, and structural arrangements, which are jointly considered in our grouping measure.

Author(s): Zhaoliang Lun, University of Massachusetts Amherst
Changqing ZOU, Simon Fraser University
Haibin Huang, University of Massachusetts Amherst, UMASS
EVangelos Kalogerakis, UMASS, University of Massachusetts Amherst
Ping Tan, Simon Fraser University
Marie Paule Cani, University of Grenoble INP
Hao(Richard) Zhang, Simon Fraser University

Speaker(s): Changqing Zou, Simon Fraser University

Summary: We propose a novel method for retrieving and placing complementary parts for assembly-based modeling tools. Our method does not require consistent segmentation and part labeling, and it can learn from a non-curated and inconsistent oversegmentation of shapes in an online repository.

Author(s): Minhyuk Sung, Stanford University
Hao Su, Stanford University
Vladimir G. Kim, Adobe Research
Siddhartha Chaudhuri, IIT Bombay
Leonidas Guibas, Stanford University

Speaker(s): Minhyuk Sung, Stanford University

Summary: We introduce a method for learning a model for the mobility of parts in 3D objects. Our method allows not only to understand the dynamic functionalities of one or more parts in a 3D object, but also to apply the mobility functions to static 3D models.

Author(s): Ruizhen Hu, Shenzhen University
Wenchao Li, Shenzhen University
Oliver van Kaick, Carleton University
Ariel Shamir, IDC The Interdisciplinary Center, Israel
Hao Zhang, Simon Fraser University
Hui Huang, Shenzhen University

Speaker(s): Ruizhen Hu & Oliver van Kaick, Shenzhen University & Carleton University

Summary: We introduce an interactive authoring pipeline that uses terrain synthesizers based on Conditional Generative Adversarial Networks trained on realworld terrains and corresponding sketches. In a creative session, the artist sketches rivers and ridges, and the algorithm automatically synthesizes a realistic terrain. An erosion synthesizer can also simulate terrain evolution efficiently.

Author(s): Eric Guérin, LIRIS
Julie Digne, CNRS, LIRIS - CNRS
Eric Galin, LIRIS - CNRS
Adrien Peytavie, LIRIS - CNRS, Université Claude Bernard Lyon 1
Christian Wolf, INSA-Lyon, LIRIS - CNRS
Bedrich Benes, Purdue University
Benoît Martinez, Ubisoft Entertainment

Speaker(s): Eric Guérin, LIRIS, INSA-Lyon, CNRS