View Interpolation and Quicktime VR

1
View Interpolation and Quicktime VR

• Presenting
• View Interpolation for Image Synthesis
• Shenchang Eric Chen
• Lance Williams
• and
• QuickTime VR - An Image-Based Approach
• to Virtual Environment Navigation
• Shenchang Eric Chen

2
View Interpolation for Image Synthesis

• How to generate intermediate views using 3-d
  Morphing of nearby images

3
View Interpolation Motivation

• Traditional 3d rendering speed is dependent on
  scene complexity
• Complex scenes may render too slowly to be
  interactive
• Geometrically modeling scenes is time-consuming
  and labor intensive
• Image-based approach can save time
• Storing large numbers of intermediate views
  requires larges amounts of data
• Interpolating views on-the-fly saves space

4
View Interpolation Implementation

• Step 1
• Establish Pixel
• Correspondence
• Determine pixel location in 3-space
• Use camera transformations to determine pixel
  offset vectors
• Store offset vectors in morph map

5
View Interpolation Implementation

• Step 2
• Interpolate Correspondences
• Use linear interpolation and forward differencing
  to determine new pixel location
• Could also use quadratic, cubic etc.
  interpolation
• Small view offsets will produce better results
• Special transformations may be simplified
• Movement parallel or perpendicular to the viewing
  plane

6
View Interpolation Implementation

• Step 3
• Image Composting
• Overlaps resolved by z-buffering
• Holes filled by re-rendering or 2d interpolation
• Use more viewpoints to create fewer holes

7
View Interpolation Results
Rendered Images
Interpolated Images (middle two)
8
Viewpoint Interpolation Applications

• Virtual Reality
• Generate photorealistic interactive 3d
  walkthroughs from a limited set of basis images
• Motion Blur
• Blur fast moving objects by generating large
  numbers of intermediate images without having to
  fully render them
• Soft Shadows
• Interpolate multiple shadow buffers to simulate
  area lighting
• Image-based primitives
• No need for mathematical object description

9
Viewpoint Interpolation Tradeoffs vs.
Traditional Rendering

• Advantages
• No 3d model required
• Speed independent of scene complexity
• Generate new views from previously acquired images

• Disadvantages
• Requires per-pixel range information
• Requires precise description of camera movements
• No viewpoint-specific lighting effects

10
QuickTime VR - An Image-Based Approachto Virtual
Environment Navigation
Simulation of a camera's motions in photographic
or computer synthesised spaces
11
Quicktime VR Motivation

• Speed
• Interactive framerates are difficult to achieve
  with traditional photorealistic rendering
• Accomodation of real scenes
• Real-world details difficult to process through
  modeling and rendering without loss of quality
• Scene complexity independence
• Traditional renderers tend to bog down with more
  complex scenes

12
QuickTime VR Implementation

• Camera Rotation
• Achieved through non-linear warping of a
  cylindrical environment map
• Actual implementation not discussed
• Supports 360 degree horizontal rotation
• Less than 180 degree vertical rotation
• Camera Zooming
• Done by changing camera field of view
• Multi-resolution textures are used to speed up
  zoomed-out views

13
QuickTime VR Implementation

• Object Rotation
• Supported with a simple 2d movie of all possible
  viewpoints of the object
• Camera Movement
• Camera is only allowed to move from one
  predefined point to another
• Free-form movement possible through view
  interpolation, but not implemented in QuickTime VR

14
QuickTime VR Tradeoffs vs. Standard 3d rendering

• Advantages
• Render Time independent of scene complexity
• Ability to render photographically acquired
  real-world scenes

• Disadvantages
• Inaccurate viewpoint specific lighting
• Camera movement limited to predefined viewpoints