Welcome

Contact

Philipp Krähenbühl
353 Serra Mall
Room 286
Stanford CA 94305-9025
 
email: philkr(at)stanford.edu
 
curriculum vitae

Research

I am a Ph.D. candidate in the Department of Computer Science at Stanford University. I am working with Vladlen Koltun and the Stanford Virtual Worlds Group. My research interests are computer graphics and lately computer vision and machine learning.

Previous research:

Teaching

Publications

Philipp Krähenbühl and Vladlen Koltun
Efficient Inference in Fully Connected CRFs with Gaussian Edge Potentials
NIPS 2011 (oral, best student paper)
[PDF] [Project Page with code]
Most state-of-the-art techniques for multi-class image segmentation and labeling use conditional random fields defined over pixels or image regions. While region-level models often feature dense pairwise connectivity, pixel-level models are considerably larger and have only permitted sparse graph structures. In this paper, we consider fully connected CRF models defined on the complete set of pixels in an image. The resulting graphs have billions of edges, making traditional inference algorithms impractical. Our main contribution is a highly efficient approximate inference algorithm for fully connected CRF models in which the pairwise edge potentials are defined by linear combinations of Gaussian kernels. Our algorithm can approximately minimize fully connected models on tens of thousands of variables in a fraction of a second. Quantitative and qualitative results on the MSRC-21 and PASCAL VOC 2010 datasets demonstrate that full pairwise connectivity at the pixel level produces significantly more accurate segmentations and pixel-level label assignments.
Sergey Levine, Philipp Krähenbühl, Sebastian Thrun, Vladlen Koltun
Gesture Controllers.
Proceedings of ACM SIGGRAPH 2010 (Los Angeles, July 25-29, 2010), ACM Transactions on Graphics
[PDF] [Video]
Gesture controllers learn optimal policies to generate smooth, compelling gesture animations from speech and other optional inputs. The accompanying video presents examples of various controllers, including controllers that recognize key words, admit manual manipulation of gesture style, and even animate a character with a non-humanoid morphology.
Philipp Krähenbühl, Manuel Lang, Alexander Hornung and Markus Gross
Retargeting of Streaming Video
Proceedings of ACM SIGGRAPH Asia (Yokohama, Japan, December 16-19, 2009), ACM Transactions on Graphics
[PDF] [Video]
We present a novel, integrated system for content-aware video retargeting. A simple and interactive framework combines key frame based constraint editing with numerous automatic algorithms for video analysis. This combination gives content producers high level control of the retargeting process. The central component of our framework is a non-uniform, pixel-accurate warp to the target resolution which considers automatic as well as interactively defined features. Automatic features comprise video saliency, edge preservation at the pixel resolution, and scene cut detection to enforce bilateral temporal coherence. Additional high level constraints can be added by the producer to guarantee a consistent scene composition across arbitrary output formats. For high quality video display we adopted a 2D version of EWA splatting eliminating aliasing artifacts known from previous work. Our method seamlessly integrates into postproduction and computes the reformatting in realtime. This allows us to retarget annotated video streams at a high quality to arbitary aspect ratios while retaining the intended cinematographic scene composition. For evaluation we conducted a user study which revealed a strong viewer preference for our method.
Philipp Krähenbühl, Manuel Lang and Markus Gross
Art Directable Retargeting for Streaming Video
Bachelor Thesis, ETH Zurich, February 2009
[PDF]
We present a novel framework for content-aware and art-directable video retargeting. A simple and interactive workflow combines key frame based constraint editing with numerous automatic algorithms for video analysis. This combination gives content producers high level control of the retargeting process. The central component of our framework is a non-uniform, pixelaccurate warp to the target resolution which considers automatic as well as interactively defined features. Automatic features comprise video saliency, edge preservation at the pixel resolution, and a scene cut detection to enforce bilateral temporal coherence. Additional high level constraints concerning scene composition can be added by the producer. For high quality video display we adopted a 2D version of EWA splatting eliminating aliasing artifacts known from previous methods. Our method seamlessly integrates into postproduction and computes the reformatting in realtime. This allows us to retarget annotated video streams to arbitary devices while retaining the intended cinematographic scene composition.