By Rob Knies, Managing Editor, Microsoft Research
In these days of nearly ubiquitous access to digital cameras, and the consequent explosion of images available via the Internet, it has become more important than ever to find ways to render those images efficiently.
Unfortunately, the complex nature of digital imagery consumes prodigious amounts of bandwidth and memory. How can resources be utilized efficiently while maintaining a high level of quality in coping with the deluge of photos accessible on the Web these days?
Spotlight: Blog post
Eureka: Evaluating and understanding progress in AI
How can we rigorously evaluate and understand state-of-the-art progress in AI? Eureka is an open-source framework for standardizing evaluations of large foundation models, beyond single-score reporting and rankings. Learn more about the extended findings.
Hugues Hoppe has an answer.
Hoppe, a principal researcher in the Graphics Group within Microsoft Research Redmond, has co-authored a paper entitled Factoring Repeated Content Within and Among Images, along with colleagues Huamin Wang of the Georgia Institute of Technology, and Yonatan Wexler and Eyal Ofek of Microsoft.
The paper is one of 13 from Microsoft Research to be accepted for SIGGRAPH 2008, the Association for Computer Machinery’s annual conference for its Special Interest Group on Graphics and Interactive Techniques.
The Microsoft Research contribution represents 14 percent of the total of 90 papers to be presented during SIGGRAPH 2008, scheduled in Los Angeles from Aug. 11 to 15. Ten of Microsoft Research’s papers were co-written by academic partners, and those papers come from the organization’s labs in Asia, Cambridge, U.K., and Redmond.
The collaboration between Hoppe, Wang, Wexler, and Ofek—the latter two of whom do research for Microsoft Virtual Earth—uses a novel technique to take advantage of shared elements of images.
“Many Web services, such as Live Search Maps, Virtual Earth, and Photosynth—and equivalents by our competitors—provide increasingly realistic and useful descriptions of the world,” Hoppe says. “However, the resulting glut of imagery data presents a real challenge for limited network bandwidth and memory capacities. Our goal is to exploit the tremendous sharing of imagery features, both within individual images and across collections of images. For instance, building façades often contain repeated elements, such as windows and bricks. And multiple images often reveal the same surfaces, only just slightly deformed. Our approach is to factor these elements efficiently so as to represent them only once.
“In a sense, our technology can be seen as orthogonal to traditional image compression such as JPEG. While JPEG looks for correlation at a local scale in the image, such as areas of constant color, we look for correlation of features at large scales. We think that this an exciting technological direction that will enable the representation of complicated 3-D environments on affordable devices.”
Other Microsoft Research projects to be presented include one from Microsoft Research Asia on a new programming language for general-purpose computation on a graphics-processing unit, work to enhance the functionality of Photosynth, and new ways to treat video to enable useful editing tasks.
Microsoft Research Asia was heavily represented, with eight paper, six co-written by Baining Guo, assistant managing director of the Beijing-based lab, and five co-written by Kun Zhou, lead researcher in the Internet Graphics Group.
Papers from Microsoft Research to be presented during SIGGRAPH 2008:
BSGP: Bulk-Synchronous GPU Programming Qiming Hou, Tsinghua University; Kun Zhou, Microsoft Research Asia; and Baining Guo, Microsoft Research Asia
Edge-Preserving Decompositions for Multi-Scale Tone and Detail Manipulation Zeev Farbman, The Hebrew University of Jerusalem; Raanan Fattal, The Hebrew University of Jerusalem; Dani Lischinski, The Hebrew University of Jerusalem; and Richard Szeliski, Microsoft Research Redmond
Example-Based Dynamic Skinning in Real-Time Xiaohan Shi, Zhejiang University; Kun Zhou, Microsoft Research Asia; Yiying Tong, Michigan State University; Mathieu Desbrun, California Institute of Technology; Hujun Bao, Zhejiang University; and Baining Guo, Microsoft Research Asia
Factoring Repeated Content Within and Among Images Huamin Wang, Georgia Institute of Technology; Yonatan Wexler, Microsoft; Eyal Ofek, Microsoft and Hugues Hoppe, Microsoft Research Redmond
Finding Paths Through the World’s Photo Noah Snavely, University of Washington; Rahul Garg, University of Washington; Steven M. Seitz, University of Washington; and Richard Szeliski, Microsoft Research Redmond
Interactive Relighting of Dynamic Refractive Objects Xin Sun, Microsoft Research Asia; Kun Zhou, Microsoft Research Asia; Eric Stollnitz, Microsoft Research Redmond; Jiaoying Shi, Zhejiang University; and Baining Guo, Microsoft Research Asia
Inverse Texture Synthesis Li-Yi Wei, Microsoft Research Asia; Jianwei Han, Zhejiang University; Kun Zhou, Microsoft Research Asia; Hujun Bao, Zhejiang University; Baining Guo, Microsoft Research Asia; and Heung-Yeung Shum, Microsoft Research Asia
Modeling Anisotropic Surface Reflectance with Example-Based Microfacet Synthesis Jiaping Wang, Microsoft Research Asia; Shuang Zhao, Shanghai Jiaotong University; Xin Tong, Microsoft Research Asia; John Snyder, Microsoft Research Redmond; and Baining Guo, Microsoft Research Asia
Parallel Poisson Disk Sampling Li-Yi Wei, Microsoft Research Asia
Progressive Inter-scale and Intra-scale Non-blind Image Deconvolution Lu Yuan, The Hong Kong University of Science and Technology; Jian Sun, Microsoft Research Asia; Long Quan, The Hong Kong University of Science and Technology; and Heung-Yeung Shum, Microsoft Research Asia
Real-Time Smoke Rendering Using Compensated Ray Marching Kun Zhou, Microsoft Research Asia; Zhong Ren, Microsoft Research Asia; Stephen Lin, Microsoft Research Asia; Hujun Bao, Zhejiang University; Baining Guo, Microsoft Research Asia; and Heung-Yeung Shum, Microsoft Research Asia
Streaming Multigrid for Gradient-Domain Operations on Large Images Michael Kazhdan, Johns Hopkins University; and Hugues Hoppe, Microsoft Research Redmond
Unwrap Mosaics: A New Representation for Video Editing Alex Rav-Acha, Weizmann Institute of Science; Pushmeet Kohli, Microsoft Research Cambridge; Andrew Fitzgibbon, Microsoft Research Cambridge; and Carsten Rother, Microsoft Research Cambridge
Graphics and multimedia