Appearance Models for Computer Graphics and Vision


Lectures/Notes/Readings

Instructor


Time and Place

Wed, 3:00 - 6:00 PM, Room 320 Soda


Course Announcement

The appearance of the everyday world has long been a topic of interest to many people from painters to physicists. Even simple questions require careful thought. Why is the sky blue? Why does wet sand look darker than dry sand? How can you reproduce a human face using oil paints? More recently appearance models have become increasingly important in computer graphics and vision. In graphics, they are needed to model and simulate different materials. In vision, texture and reflection models can be used to guide the acquisition of computer models of different scenes and objects, as well as the recognition of these scenes and objects in images.

This class is the second part of a unique two course sequence being taught jointly at Stanford and Berkeley. The first course was offered at Stanford in Fall 2000 and covered methods for measuring reflection, models of reflection from rough surfaces, subsurface reflection, and light interaction with participating media and atmospheric models. The second course will be at Berkeley this Spring semester (Winter quarter in the Stanford calendar). It will cover human perception of shape, illumination, reflectance and texture as well as computational models for recovering these attributes of a scene from a set of images.

The class is open to students with a background in computer graphics, computer vision or human vision. Even though this is the second course in a sequence, the first course is not a prerequisite.

In addition to learning specifically about appearance modeling, students may expect to gain a uniquely interdisciplinary perspective on the relationship between computer vision, computer graphics and human vision. Typically, the output of computer graphics is pictures for humans to see--hence knowing the characteristics of human vision enables concentrating on what matters. Computer vision can be used to acquire models of scenes for use in computer graphics; and graphics rendering can in turn serve as a test for the adequacy of these models. The course will emphasize these connections.

List of topics:
  1. Overview of human vision
  2. Brightness and lightness perception; Tone mapping
  3. Shape from shading: psychophysics and models
  4. Perception of specularities and shadows
  5. Inverse global illumination
  6. General considerations on solution of inverse problems
  7. Texture perception, analysis and synthesis
  8. Generative models for texture
  9. Human faces
  10. Human movement


Lectures, Notes, and Readings
Jan 17, 2001
Jitendra Malik
Front End Visual Processing: Adaption, Filtering, Sensitivity, Hyperacuity, perceptual organization, surface attitude perception, visual systems as photometer/geometer. (PowerPoint notes)
Jan 24, 2001
David Forsyth
Colour
Jan 31, 2001
Jitendra Malik
Perception of Brightness, Lightness, Transparency. Tone mapping for computer graphics.
Feb 7, 2001
Jitendra Malik
Shape from Shading: Psychophysics, Computational models ignoring interreflections, Qualitative structure from photometric invariants
Feb 14, 2001
Jitendra Malik
Perception of surface shape and material from highlights. Psychophysics and computational models. Perception of Shadows: Psychophysics and computational models. Techniques of artists.
Feb 21, 2001
Pat Hanrahan & Jitendra Malik
Inverse Illumination, Illumination as Convolution, Inverse Global Illumination, Psychophysics, General remarks on inverse problems.
Feb 28, 2001
Jitendra Malik
Inverse illumination problems, Inverse Global Illumination, discussion on texture by Alyohsa.
Mar 7, 2001
Jitendra Malik
Texture, and more discussion by Alyosha
Mar 15, 2001
Jitendra Malik
Texture Cont. + Faces.
Course web site: http://www.cs.berkeley.edu/projects/vision/classes/cs294-appearance_models/sp2001/
Questions or comments --> Alex Berg at aberg@cs.berkeley.edu