• Nebyly nalezeny žádné výsledky

Automatic Creation of Object Hierarchies for Ray Tracing of Dynamic Scenes - WSCG '07 -

N/A
N/A
Protected

Academic year: 2022

Podíl "Automatic Creation of Object Hierarchies for Ray Tracing of Dynamic Scenes - WSCG '07 -"

Copied!
18
0
0

Načítání.... (zobrazit plný text nyní)

Fulltext

(1)

Automatic Creation of Object Hierarchies for Ray Tracing of Dynamic Scenes

- WSCG '07 -

Martin Eisemann, Thorsten Grosch, Marcus Magnor, Stefan Müller

Computer Graphics Lab,

TU Braunschweig, Germany Institute for Computational Visualistics,

eisemann@cg.tu-bs.de

(2)

Motivation

Ray Tracing

Interactive or real-time frame rates possible

[Parker et al. 99], [Wald 04], [Reshetov et al. 05]

Strongly dependent on acceleration data structures

Optimized for static scenes

(3)

Ray Tracing of Dynamic Scenes

Related Work:

[Glassner 88], [Reinhard et al. 00], [Lext et al. 01], [Günther et al. 06], [Lauterbach et al. 06],

[Wächter et al. 06], [Wald et al. 06]

2 Methods for dynamic scenes:

Dynamic Goldsmith and Salmon

, see also [Goldsmith and Salmon 87]

Loose Bounding Volume Hierarchy

, see also [Ulrich 00]

(4)

Dynamic Goldsmith and Salmon

Goals:

Exploitation of localities

Prevention of thinning

Initial build:

Minimize amount of expected intersection tests

Use SAH, Median-Cut, G. & S., ...

(5)

Dynamic Goldsmith and Salmon

Exploitation of localities

(6)

Dynamic Goldsmith and Salmon

Exploitation of localities

1. Deletion

(7)

Dynamic Goldsmith and Salmon

Exploitation of localities

1. Deletion

2. Find new insertion node

Insertion node

(8)

Dynamic Goldsmith and Salmon

Exploitation of localities

1. Deletion

2. Find new insertion node 3. Reinsertion

Insertion node

(9)

Dynamic Goldsmith and Salmon

Thinning

Number of objects in a node decreases

Surface area stays constant

Needs quality criterion

Initial calculation of Q(N) for every node

(10)

Dynamic Goldsmith and Salmon

Thinning

If threshold is exceeded during animation:

1. Delete node

2. Reinsert child nodes

(11)

Loose Bounding Volume Hierarchy

Reconstruction in O(n)

Hybrid between spatial acceleration data structure and BVH

Spatial median-cut with alternating axes

User-defined depth of 3N

(12)

Loose Bounding Volume Hierarchy

Lowest level of BVH is a pseudo-uniform grid

Resolution 2N x 2N x 2N

...

...

... ...

...

(13)

Loose Bounding Volume Hierarchy

Wide Object Isolation

...

...

...

...

(14)

Loose Bounding Volume Hierarchy

Skip Indices

...

...

...

...

Skip Index

(15)

Loose Bounding Volume Hierarchy

Refitting by backward iteration

Adjusts BVs

Sets skip indices

Marks empty nodes

1 2 3 4 5 6 7 ... n

Refitting-order

BVH

(16)

Test Results

Test scenes:

between 5 and 149.058 animated objects

(17)

Test Results

Dynamic G. & S.

Local movement

up to a few

hundred objects

Good overall performance

Loose BVH

Fast and constant updates

Several thousand objects

Teapot in the

stadium problem

Update-Phase RT-Phase Update-Phase RT-Phase 6x – 103x 1.0x – 1.9x speed-up 11.2x – 18.5x 0.5x – 7.0x 17ms – 907ms 6.0s – 11.4s avg. timings 125ms – 404ms 3.2s – 11.7s

(18)

End

http://graphics.tu-bs.de/people/eisemann

Odkazy

Související dokumenty

◼ The BRDF is a model of the bulk behavior of light on the microstructure when viewed from

We have demonstrated an n-gon (triangle, quad, tri- angle fan, triangle strip) scene compression algorithm which can compress such a scene in linear O(n) time with constant O(1)

Determining the topology of intersection curves is one of the important issues of surface-surface intersection problem used in Computer Aided Geometric Design and Computer

– Assign A to closer child node -> R, – Assign C to farther child node -> G, – Assign max(A,B) to parent node -> B, – Assign max(A,B,C,D) as silhouette for. parent

Thanks to this new holistic approach to expressing scenes for computer graphics and the dedicated rendering pipeline, it was possible to process and render a potentially

➨ při výpočtu stínovacího paprsku beru v úvahu jen tělesa zaznamenaná v buňce směrové krychle pro.

First, we run feature matching for all pairs of images, construct n-view matches, and group the n-view matches into motion groups.. Second, we select one image pair to initialize

Department: Department of Computer Graphics and Interaction Thesis reviewer: Václav Rechtberger.. Reviewer’s