Gamasutra

If you missed SIGGRAPH 2017 watch a selection of recorded Live Streaming Sessions.

You are here

Technical Papers

Get More Out of Your Photo

HP VR Backpack

ACM DL

Monday, 31 July, 3:45 pm - 5:35 pm, Los Angeles Convention Center, Los Angeles Convention Center - Room 150/151
Session Chair: George Drettakis, Inria, Université Côte d'Azur

VNect: Real-Time 3D Human-Pose Estimation With a Single RGB Camera

This real-time, CNN-based method for full global 3D human-body pose estimation from a single RGB camera leverages a novel fully convolutional pose formulation in conjunction with a kinematic skeleton-fitting method to yield temporally consistent 3D-pose reconstructions comparable in quality to RGB-D approaches.

Dushyant Mehta
Max Planck Institut für Informatik, Universität des Saarlandes

Srinath Sridhar
Max Planck Institut für Informatik

Oleksandr Sotnychenko
Max Planck Institut für Informatik

Helge Rhodin
Max Planck Institut für Informatik

Mohammad Shafiei
Universität des Saarlandes, Max Planck Institut für Informatik

Hans-Peter Seidel
Max Planck Institut für Informatik

Weipeng Xu
Max Planck Institut für Informatik

Dan Casas
Universidad Rey Juan Carlos

Christian Theobalt
Max Planck Institut für Informatik, Max Planck Center for Visual Computing and Communication

Real-Time Geometry, Albedo, and Motion Reconstruction Using a Single RGBD Camera

A real-time system that can simultaneously reconstruct a casual scene with a detailed geometry model, surface albedo, per-frame non-rigid motion and environmental lighting using a single RGBD camera.

Kaiwen Guo
Tsinghua University

Feng Xu
Tsinghua University

Tao Yu
Beihang University

Xiaoyang Liu
Tsinghua University

Qionghai Dai
Tsinghua University

Yebin Liu
Tsinghua University

Modeling Surface Appearance From a Single Photograph Using Self-Augmented Convolutional Neural Networks

A novel CNN-based method for modeling physically plausible surface appearance from a single photograph of a planar sample under unknown and uncontrolled natural lighting. To reduce the required size of the labeled training dataset, a novel augmented CNN training strategy leverages unlabeled data during training.

Xiao Li
Microsoft Research Asia, University of Science and Technology of China

Yue Dong
Microsoft Research Asia

Pieter Peers
College of William & Mary

Xin Tong
Microsoft Research

Virtual Rephotography: Novel View Prediction Error for 3D Reconstruction

This paper evaluates end-to-end image-based modeling and rendering systems by analyzing how well they predict secret test images. The technique can be used, for example, to locally highlight 3D reconstruction errors.

Michael Waechter
Technische Universität Darmstadt

Mate Beljian
Technische Universität Darmstadt

Simon Fuhrmann
Technische Universität Darmstadt

Nils Moehrle
Technische Universität Darmstadt

Johannes Kopf
Facebook, Inc.

Michael Goesele
Technische Universität Darmstadt

Computational Zoom: A Framework for Post-Capture Image Composition

Introducing computational zoom, a framework that allows a photographer to manipulate several aspects of composition in post-processing from a stack of pictures captured at different distances from the scene. Using this framework, the photographer can change the relative magnification and perspective of objects at different depths in post-processing.

Abhishek Badki
University of California, Santa Barbara

Orazio Gallo
NVIDIA Research

Jan Kautz
NVIDIA Research

Pradeep Sen
University of California, Santa Barbara