Machine Learning for PC Design A brief presentation.


108 views
Uploaded on:
Category: Sales / Marketing
Description
Machine Learning for PC Illustrations A brief presentation. By Dr. Zhang Hongxin State Key Lab of CAD&CG, ZJU . Plot. Foundation What is Machine Realizing? Is it truly valuable for PC illustrations? Our arrangement. The biggest test of Today's CG&A.
Transcripts
Slide 1

Machine Learning for Computer Graphics A brief presentation By Dr. Zhang Hongxin State Key Lab of CAD&CG, ZJU

Slide 2

Outline Background What is Machine Learning? Is it truly valuable for PC representation? Our arrangement

Slide 3

The biggest test of Today’s CG&A The repetitive exertion needed to make advanced universes and computerized life. “Finding better approaches to impart and new sorts of media to create.” Filmmakers, researchers, visual originators, fine specialists, and diversion planners.

Slide 6

Pure procedural combination versus Unadulterated information Creating movements for a character in a motion picture Pure procedural amalgamation. smaller, yet exceptionally fake, seldom use by and by. “By hand” or “pure data”. higher quality yet lower adaptability. the best of both universes: half and half techniques?!?

Slide 7

Bayesian Reasoning Principle demonstrating of instability. Broadly useful models for unstructured information. Compelling calculation for information fitting and investigation under vulnerability. However, right now it is constantly utilized as a black box.

Slide 8

Data driven displaying

Slide 9

What is machine learning? Manmade brainpower Statistics and Bayesian techniques Machine Learning Data mining (KDD) Control and data Theory Data-base Computer Vision Multi-media Bio-informatics Computer Graphics

Slide 10

What is machine learning? (Cont.) Definition by Mitchell, 1997 A system gains for a fact E as for some class of errands T and execution measure P, if its execution at undertaking T, as measured by P, enhances with experience E. Learning frameworks are not specifically modified to take care of an issue, rather build up own system in light of: illustrations of how they ought to act from experimentation experience attempting to tackle the issue Different than standard CS: need to actualize obscure capacity, just have admittance to test info yield sets (preparing samples) Hertzmann, 2003 For the reasons of PC representation, machine learning should be seen as an arrangement of strategies for utilizing data(???) .

Slide 11

Why Study Learning? Create upgraded PC frameworks naturally adjust to client, redo frequently hard to secure vital learning find designs disconnected from the net in huge databases ( information mining ) Improve comprehension of human, organic learning computational examination gives solid hypothesis, forecasts blast of strategies to break down cerebrum movement amid learning Timing is great developing measures of information accessible shoddy and effective PCs suite of calculations, hypothesis officially created

Slide 12

Main class of learning issues Learning situations contrast as per the accessible data in preparing illustrations Supervised : right yield accessible Classification : 1-of-N yield (discourse acknowledgment, object recognition,medical determination) Regression : genuine esteemed yield (anticipating business sector costs, temperature) Unsupervised : no criticism, need to build measure of good yield Clustering : Clustering alludes to systems to fragmenting information into sound “clusters.” Reinforcement : scalar input, perhaps transiently deferred

Slide 13

And more … Time arrangement investigation. Measurement lessening. Model determination. Nonexclusive routines. Graphical models.

Slide 14

Is it truly helpful for PC illustrations? Con: Everything is machine learning or everything is human tuning? Here and there, this may be valid. Ace: all the more comprehension of adapting, yet yields a great deal all the more capable and viable calculations. Issue scientific categorization. Broadly useful models. Prevailing upon probabilities. I trust the mathematic enchantment.

Slide 15

Mesh Processing– grouping/division Hierarchical Mesh Decomposition utilizing Fuzzy Clustering and Cuts. By Sagi Katz and Ayellet Tal, SIGGRAPH 2003

Slide 16

Texture combination and investigation - Hidden Markov Model " Texture Synthesis over Arbitrary Manifold Surfaces", by Li-Yi Wei and Marc Levoy. In Proceedings of SIGGRAPH 2001. "Fast Texture Synthesis utilizing Tree-organized Vector Quantization", by Li-Yi Wei and Marc Levoy. In Proceedings of SIGGRAPH 2000.

Slide 17

Image handling and amalgamation by graphical Model Image Quilting for Texture Synthesis and Transfer. Alexei A. Efros and William T. Freeman. SIGGRAPH 2001. Graphcut Textures: Image and Video Synthesis Using Graph Cuts. Vivek Kwatra, Irfan Essa, Arno Schã¶dl, Greg Turk, Aaron Bobick. SIGGRAPH 2003.

Slide 19

BTF – reflectance composition union Synthesizing Bidirectional Texture Functions for Real-World Surfaces. Xinguo Liu, Yizhou Yu and Heung-Yeung Shum. Later papers…

Slide 20

Style machines - Time arrangement investigation By Matthew Brand (MERL) and Aaron Hertzmann. SIGGRAPH 2000

Slide 21

Motion surface - straight element framework Yan Li, Tianshu Wang, and Heung-Yeung Shum. Movement Texture: A Two-Level Statistical Model for Character Motion Synthesis.

Slide 22

Video Textures – Reinforce Learning Arno Schã¶dl, Richard Szeliski, David H. Salesin, and Irfan Essa.â  Video surfaces .  Proceedings of SIGGRAPH 2000 , pages 489-498, July 2000.

Slide 23

Human shapes – Dimension Reduction The Space of Human Body Shapes: Reconstruction and Parameterization From Range Scans. Brett Allen, Brian Curless, Zoran Popoviä‡. SIGGRAPH 2003. A Morphable Model for the Synthesis of 3D Faces. Volker Blanz and Thomas Vetter. SIGGRAPH 1999.

Slide 24

Our Plan The SIG-ML4CG http://www.cad.zju.edu.cn/home/zhx/ML4CG/

Slide 25

Schedule

Slide 26

Textbooks Information Theory, Inference, and Learning Algorithms, by David MacKay. Machine Learning by Tom Mitchell. Information mining: Concepts and Techniques By Jiawei Han and Micheline Kamber Pattern Classification (second ed.) by Richard O. Duda, Peter E. Hart and David G.

Slide 27

Reference Links ML4CG by Hertzmann http://www.dgp.toronto.edu/~hertzman/mlcg2003/hertzmann-mlcg2003.pdf DDM ideas http://datamining.ihe.nl/symposium/intro.htm .

Recommended
View more...