320 likes | 438 Vues
This collection highlights groundbreaking research by leading experts in graphics, vision, and HCI, including Li-Yi Wei, Wenping Wang, and Kenneth Wong. Topics range from real-time cloth animation methods in video games to advanced techniques in image processing and computational photography. The researchers emphasize a data-driven approach, exploring natural repetitions and procedural generation. With contributions to major conferences like SIGGRAPH, the work exemplifies pushing boundaries in computer graphics while ensuring an engaging and enjoyable learning experience for students.
E N D
Graphics, Vision, HCI • K.P. Chan • Wenping Wang • Li-Yi Wei • Kenneth Wong • Yizhou Yu
Li-Yi Wei • Background • Stanford (95-01), NVIDIA (01-05), MSR (05-11) • Research • Nominal: Graphics, HCI, parallelism • Actual: Computing natural repetitions • (Computer science is about repetitions) • Can work on almost anything + have fun • I tailor projects for individual students (so they also have fun)
Computing natural repetitions data driven (non-parametric) motion texture inverse synthesis HDR edit texture synthesis element texture auto interact parallelism graphics HCI blue noise parallel random Parallel Poisson differential analysis procedural (parametric) revision control
Discrete element textures[Ma et al. SIGGRAPH 2011] exemplar synthesis domain output
SIGGRAPH • The coolest (& ass kicking) venue in graphics • Each paper can be worth a PhD thesis • (Just in case you don’t know) • HKU has 4 papers in SIGGRAPH 2012 • So we are awesome (in addition to have fun)
output input
output input
output input
Yizhou Yu • Background • Berkeley (PhD 2000), UIUC (2000 - 2010) • Research • Graphics, vision, image processing • Computational Photography • Computer Animation • Geometry Processing • Medical Imaging • Video Analytics
Deformation transfer for real time cloth animation [SIGGRAPH 2010] Deformation Transformer
Motivation • Real-Time Cloth Animation • Video games, virtual fashion, etc. • The Problem • Real-time performance on high-resolution models • PDE Integration, Collision resolution. Final Fantasy XIII Nurien
Overview • Hybrid Approach : • Simulate low-res cloth on the GPU • Rely on a data-driven model to transform the low-res simulation into a high-res animation Deformation Transformer
An Example High-Res Dress: 27K Triangles, Low-Res Dress: 200 Triangles Frame Rate: 261
Data-Driven Image Color Theme Enhancement [SIGGRAPH Asia 2010] Photo Reuse: how to edit a photograph to enhance a desired color impression by exploiting prior knowledge extracted from an existing photo collection? Waiting for the right season and illumination could be extremely time-consuming! source image nostalgic lively
Our Goal Image Color Theme Enhancement desolate Input Image lively
Results happy sad spring in the air peaceful Input Images
Wenping Wang • Background • Alberta (PhD 1992), Department Head • Research • Computer graphics • Geometry Processing • Computational geometry • Architectural Design • Scientific Visualization
Kwan-Yee Kenneth Wong • Background • Cambridge (PhD 2001) • Research • 3D modeling • Video surveillance • Image processing • Pattern recognition • …
contour generator N silhouette 3D Model Reconstruction • Robust recovery of shapes with unknown topology from the dual space (PAMI 2007)
dual surface original surface 3D Model Reconstruction • Robust recovery of shapes with unknown topology from the dual space (PAMI 2007) tangent operation tangent operation original surface
3D Model Reconstruction • Robust recovery of shapes with unknown topology from the dual space (PAMI 2007)
Eye Gaze Tracking • Reconstruction of display and eyes from a single image (CVPR 2010)
Eye Gaze Tracking • Reconstruction of display and eyes from a single image (CVPR 2010)
Kwok-Ping Chan • Background • HKU (PhD 1989) • Research • To apply various Machine Learning methods on Pattern Recognitions, such as facial expression recognition. • Study on Cross Domain Learning where the training and the testing domain are not the same.
Facial Expression Recognition Goal: to recognize one of the seven basic facial expressions:
Methods • Dynamic Bayesian Network • Discriminative Hidden Markov Models • Discriminative Temporal Topic Models • Given an image sequence of facial expression, we compute the probability of each expression using the above techniques.
Examples: Smile with blinking eyes: • From input, produce output • similar to input • arbitrary size Key Publication: CVPR 2009