Shape Analysis

Monday, 25 July, 3:45 pm - 5:35 pm, Anaheim Convention Center, Ballroom E
Session Chair: Misha Kazhdan, The Johns Hopkins University

RAID: A Relation-Augmented Image Descriptor

This relation-augmented image descriptor supports queries based on relations between image regions, like a person riding object X, or a plank bridging two objects. RAID successfully extracts non-trivial images, demonstrating complex inter-region relations that are easily missed or erroneously classified by existing methods.

Paul Guerrero
University College London, King Abdullah University Of Science And Technology

Niloy Mitra
University College London

Peter Wonka
King Abdullah University Of Science And Technology

Learning How Objects Function via Co-Analysis of Interactions

Introducing a co-analysis method that learns a functionality model for an object category, (for example, strollers or backpacks). The co-analysis localizes the studied properties to the specific locations, or surface patches, that support specific functionalities of the shapes, enabling functionality-aware modeling applications.

Ruizhen Hu
Shenzhen University

Oliver van Kaick
Carleton University

Bojian Wu
Shenzhen Institute of Advanced Technology

Hui Huang
Shenzhen University, Shenzhen Institute of Advanced Technology

Ariel Shamir
Interdisciplinary Center Herzliya

Hao Zhang
Simon Fraser University

PATEX : Exploring Pattern Variations

PATEX is a method to characterize and efficiently identify variations of a 2D pattern. Given a pattern and a user edit, it facilitates exploration of the space of possible pattern variations.

Paul Guerrero
University College London

Gilbert Bernstein
Stanford University

Wilmot Li
Adobe Research

Niloy Mitra
University College London

3D Mesh Labeling via Deep Convolutional Neural Networks

Experiments on multiple public benchmarks show that this novel approach for 3D mesh labeling via deep convolutional neural networks works robustly and outperforms state-of-the-art systems.

Kan Guo
Beihang University

Dongqing Zou
Beijing Samsung R&D Center

Xiaowu Chen
BeiHang University

Efficient 3D Object Segmentation From Densely Sampled Light Fields With Applications to 3D Reconstruction

An efficient algorithm to automatically segment static foreground objects from cluttered backgrounds in light fields. The method exploits high spatio-angular sampling, which increases coherence in the data and reveals new structures.

Kaan Yücer
The Walt Disney Company, ETH Zürich

Alexander Sorkine-Hornung
Disney Research

Oliver Wang
Adobe Research

Olga Sorkine-Hornung
ETH Zürich