Scribd
Upload
9 views

Parallel Distributed, GPU-Accelerated, Advanced Lighting Calculations

Uploaded by

santhosh
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
9 views

Parallel Distributed, GPU-Accelerated, Advanced Lighting Calculations

Uploaded by

santhosh
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 19

Parallel Distributed, GPU-Accelerated,

Advanced Lighting Calculations for Large-


Scale Volume Visualization

Presenter : Korlepara Sai Phanindra


M.Tech CSA
IISc Bangalore
Motivation

• Advancements in Global Illumination Optimizations.


• Limitations in said Optimizations.
• Recent advancements in data acquisition methods.
• Challenges with Distributed Compute networks
• Main Goals for the paper

2
Related Work

• Solving the Radiative Transfer Equation (RTE)


• Local Illumination methods.
• Photon tracing methods.
• Slicing / Sweeping methods.
• 3D – Summed area Tables (SATs).
• Numerical methods

3
Distributed Volumetric Illumination

Solving the RTE

• Background
• Emitted Radiance
• In-scattered Radiance

• Phase function determines the angle dependence

4
Approximating In-scattered Radiance

Modifying the sphere with cone model

• Assumed the presence of Forward scattering phase function


• P’ is a normalization factor
• Wider cones imply more scattering

5
Approximation using Level sets

To avoid the need for sampling


multiple rays in the domain

• Light source intensity


• Exponential decay term
• Sampling at different points along the
ray
Determining the Level set to select

6
Clipmaps

• Octree
• Size scaling in clipmaps
• Speed of clipmaps
• Databrick processing
• Data Transfer between nodes

7
Pre-Filtered Classification

Modifications to Transfer Function for interactive rendering

• Opacity value at any given distance based on the level set calculation
• S – the scalar value in the field, alpha the opacity value
• Scalar values cannot be stored during the duration of rendering
• Change in the Transfer function once opacity is calculated is a
challenge - A
• Approximating the scalar distribution as a gaussian for each voxel

• Further approximation can be done by discrete integration.


• Creation of Lookup tables for transfer function of different mu, sigmas.

8
Implementation

• Mean and deviation are calculated at each level set and distributed.
• Clipmap stored inside a 3D Texture, LoDs stacked on z-axis.
• Down sampling the original data brick to save on GPU memory.
Lighting Calculation
• Image space VCT : Cones Traced at each sampling point on the
viewing ray
• Object space VCT : Precomputed cones stored in light volume for each
voxel. Lower visual quality.
• As a compromise : first few levels in the cone are sampled similar to
image space, rest blended with stored in light volume.
• Light volumes are usually lower resolution than the base datagrid.
• Adaptive sampling along the Voxel cone Ray.
Surface shading : Blinn-Phong

9
Results and Discussion
Supernova Dataset

11
Boussinesq flow Dataset – 4096 cubed

12
Supernova Dataset

13
Comparative Results

14
Performance Evaluation – Strong scaling

15
Performance Evaluation – Weak scaling

16
Performance Evaluation – pre processing time

17
Performance Evaluation – Transfer Function
Modification

18
Thank You​

ClickQuestions?
to edit master
text style

You might also like