Layered Depth Images - PowerPoint PPT Presentation

1 / 19
About This Presentation
Title:

Layered Depth Images

Description:

Splat is an alpha Gaussian. Size determines which level of the Octree. Building an LDI Tree ... Calculate splat. Recursively descend octree to appropriate res ... – PowerPoint PPT presentation

Number of Views:223
Avg rating:3.0/5.0
Slides: 20
Provided by: chr1224
Category:
Tags: depth | images | layered | splat

less

Transcript and Presenter's Notes

Title: Layered Depth Images


1
Layered Depth Images
2
Choosing Sprites
  • Planar Sprites
  • Sprites with Depth
  • Layered Depth Images
  • Polygonal Sprites

3
Layered Depth Images
  • DepthPixel
  • RGBA, Z
  • Depth Image
  • Camera, DepthPixelwidth,height
  • LayeredDepthPixel
  • numLayers, DepthPixelnumLayers
  • LayeredDepthImage
  • Camera, LayeredDepthPixelwidth,height

4
Generating LDIs
  • From a model
  • Depth Images, Modified Ray Tracing
  • From the real world
  • Correspondence, Laser Range Finder

5
Rendering LDIs
  • Transformation for novel views
  • C1 and C2 are the two camera matrices
  • Splatting
  • Size d1/d2
  • Alpha approximate Gaussian

6
LDI Results
  • Time
  • Pentium II, 300 MHz, 320 MB ram
  • 320x320 pixels
  • 8-10 frames/second
  • By Moores Law 256-320 frames/second today
  • Pretty pictures

7
Layered Depth Images
8
LDI Demo
9
LDI Tree
  • Improve sampling resolution
  • Progressive Refinement
  • Gap filling
  • Interested in accuracy not run-time

10
What is an LDI tree?
  • Octree
  • Tree-nodes
  • BoundingBox
  • Children8
  • LayeredDepthImage

11
Building an LDI Tree
  • Same Sources
  • Which tree node gets each pixel
  • PixelCenter
  • C is the cameras center of projection in World
    space
  • a and b are the basis vectors for the image plane
  • c is the viewing direction
  • d is 1/z (the distance from the viewing plane)
  • SplatSize
  • Splat is an alpha Gaussian
  • Size determines which level of the Octree

12
Building an LDI Tree
  • Unfiltered pixels
  • Pixel placed in scale appropriate tree-node
  • Filtered pixels
  • Pixel is splatted into all ancestor node LDIs

13
Building an LDI Tree
  • For each reference image
  • For each pixel
  • Calculate pixel center
  • Calculate splat
  • Recursively descend octree to appropriate res
  • Splat filtered pixel into each LDI
  • Splat unfiltered pixel into appropriate res
    tree-node

14
Rendering an LDI tree
  • Recursive Render
  • Render (Tree-Node)
  • 1. If outside of view frustum return
  • 2. Estimate size of the splats
  • 3. If need higher-res or viewer inside bounding
    box
  • 4. Call Render() recursively for each child
  • 5. Warp unfiltered pixels to the Output buffer
  • 6. else
  • 7. Warp all pixels in LDI to the output buffer

15
Rendering an LDI tree
  • Hierarchy destroys LDI depth ordering
  • Output Buffer is another LDI
  • Render Output Buffer by splatting to screen

16
Progressive Refinement
  • This should be obvious

17
Gap Filling
  • First Pass
  • Render as usual
  • Create a stencil for gaps
  • Second Pass
  • Render filtered pixels (low res version) in the
    stencil bottom up updating the stencil

18
(No Transcript)
19
LDI Tree Video
Write a Comment
User Comments (0)
About PowerShow.com