Technical Papers
Image Processing
Monday, 6 August 2:00 PM - 3:30 PM | Los Angeles Convention Center, Room 515AB
Session Chair: Peter-Pike Sloan, NVIDIA Corporation
Image Processing
Decoupling Algorithms From Schedules for Easy Optimization of Image-Processing Pipelines
A new representation for image-processing algorithms, which separates intrinsic algorithm from execution order and storage, enables clean, modular code to compile to high-efficiency implementations across many architectures. The compiler produces high-performance implementations on ARM, x86, and GPUs, targeting SIMD units, multiple cores, and complex memory hierarchies.
Jonathan Ragan-Kelley
Massachusetts Institute of Technology
Andrew Adams
Massachusetts Institute of Technology
Sylvain Paris
Adobe Systems Incorporated
Marc Levoy
Stanford University
Saman Amarasinghe
Massachusetts Institute of Technology
Frédo Durand
Massachusetts Institute of Technology
Adaptive Manifolds for Real-Time High-Dimensional Filtering
A technique for performing high-dimensional filtering of images and videos in real time. This paper presents formal derivations for the equations that define the filter and shows that it produces high-quality results, is faster, and requires less memory than previous approaches.
Eduardo S. L. Gastal
Universidade Federal do Rio Grande do Sul
Manuel M. Oliveira
Universidade Federal do Rio Grande do Sul
High-Quality Image Deblurring With Panchromatic Pixels
A high-quality image deblurring system with new image priors based on an imaging system that has a newly designed sensor pattern achieved by adding panchromatic pixels that are sensitive to all wave-lengths of visible light and collect a significantly higher proportion of the light striking the image sensor.
Sen Wang
Eastman Kodak Company
Tingbo Hou
Stony Brook University
John Border
Eastman Kodak Company
Hong Qin
Stony Brook University
Rodney Miller
Eastman Kodak Company
Practical Temporal Consistency for Image-Based Graphics Applications
This approach approximates image-based minimization problems with a fast edge and motion-aware filtering method, which allows computation of temporally consistent results over video sequences. The paper shows that this simple approach can generate high-quality stable results for optical flow, disparity estimation, data upsampling, and colorization.
Manuel Lang
Disney Research Zürich and ETH Zürich
Oliver Wang
Disney Research
Tunc Aydin
Disney Research
Aljoscha Smolic
Disney Research
Markus Gross
Disney Research Zürich and ETH Zürich