Master Thesis Real-time Realistic Rain Rendering Carles Creus - - PowerPoint PPT Presentation

Introduction State of the art Our method Results and Conclusions

Master Thesis Real-time Realistic Rain Rendering

Carles Creus

Advisor: Gustavo Patow Facultat d’Inform` atica de Barcelona July 8, 2010

Introduction State of the art Our method Results and Conclusions

1

Introduction Motivation Rain phenomena

2

State of the art Approaches Summary

3

Our method Proposal Preprocess Real-time

4

Results and Conclusions Tests Conclusions

Introduction State of the art Our method Results and Conclusions

Motivation

Weather is used to transmit specific moods Filming becomes expensive and laborious Synthetic methods simplify the task Our focus: rain rendering Difficulties Huge variety of phenomena Complex physical evolutions, optical properties Overwhelming amount of small details

Introduction State of the art Our method Results and Conclusions

Rain phenomena - Raindrops

Introduction State of the art Our method Results and Conclusions

Rain phenomena - Puddles, splashes, coronas and ripples

Introduction State of the art Our method Results and Conclusions

Rain phenomena - Other

Introduction State of the art Our method Results and Conclusions

State of the art

Each method simulates a specific subset of phenomena Simplifications on optical properties and physics Traditionally, framerate chosen over realism

Introduction State of the art Our method Results and Conclusions

Garg - 2006

Objectives

1 Complex lighting patterns for close-up shots:

Create a photorealistic model to render rain streaks using: → light direction → view direction → raindrop shape Create a database of precomputed rain streak renders

2 Add rain streaks to videos

Introduction State of the art Our method Results and Conclusions

Garg - Model

Base Oscillation model developed in atmospheric sciences: Assumes that the equilibrium shape is spherical The shape is expressed as a combination of harmonics But it does not specify the parameters for the oscillation → Capture real images and compare with synthetic renders

Introduction State of the art Our method Results and Conclusions

Garg - Real captures

Setup Drops released from 15m r0 = 2mm Light 1m away HDR camera 3m away 10 repetitions

Introduction State of the art Our method Results and Conclusions

Garg - Comparison

Introduction State of the art Our method Results and Conclusions

Rousseau - 2006

Introduction State of the art Our method Results and Conclusions

Tariq - 2007

Introduction State of the art Our method Results and Conclusions

Tatarchuk - 2006

Introduction State of the art Our method Results and Conclusions

Centelles - 2009

Introduction State of the art Our method Results and Conclusions

Summary

Real-time Raindrops Ground collision Moving camera Wind Lightning Reflection, Refraction Participating media Splashes Ripples Dripping Garg ✗ ✓ ✓ ✓ ✗ ✗ ✗ ✗ ✗ ✗ ✗ Wang ✓ ✓ ✗ ✓ ✗ ✗ ✗ ✗ ✗ ✗ ✗ Rousseau ✓ ✓ ✗ ✓ ✗ ✗ ✓ ✗ ✗ ✗ ✗ Tariq ✓ ✓ ✗ ✓ ✓ ✗ ✗ ✓ ✗ ✗ ✗ Centelles ✓ ✓ ✗ ✓ ✗ ✗ ✗ ✓ ✗ ✗ ✗ Tatarchuk ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Our method ✓ ✓ ✓ ✓ ✗ ✗ ✗ ✗ ✓ ✗ ✗

Introduction State of the art Our method Results and Conclusions

Summary

Open issues ✗ Comprehensive algorithms have poor user interaction ✗ Restricted artistic direction ✗ Each phenomena decoupled from the rest ✗ Seldom interaction with the scene

Introduction State of the art Our method Results and Conclusions

Proposal

Objectives Real-time rendering of raindrops Realistic raindrop illumination Interaction with the scene:

splashes no indoor rain

Arbitrary rain placement and density

Introduction State of the art Our method Results and Conclusions

Overview

Introduction State of the art Our method Results and Conclusions

Overview

Introduction State of the art Our method Results and Conclusions

Preprocess - Atlas

Introduction State of the art Our method Results and Conclusions

Preprocess - Atlas

450 images 2880 x 3606 12 mipmap levels

Introduction State of the art Our method Results and Conclusions

Preprocess - Atlas

5 images 32 x 3606 12 mipmap levels

Introduction State of the art Our method Results and Conclusions

Preprocess - Splash animation

21 frames 48 x 32 6 mipmap levels

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Rain space scheme

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Rain space scheme - Optimization

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Particle packets Worldwide Tree:

Quad-tree Kd-tree

Grid

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Particle packets Worldwide Tree:

Quad-tree Kd-tree

Grid

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Particle packets Worldwide Tree:

Quad-tree Kd-tree

Grid

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Particle packets Worldwide Tree:

Quad-tree Kd-tree

Grid

Introduction State of the art Our method Results and Conclusions

Preprocess - Particle generation

Particle packets Worldwide Tree:

Quad-tree Kd-tree

Grid

Introduction State of the art Our method Results and Conclusions

Real-time - CPU - Time animation

Fall animated with a global parameter in [0, 1). Updated with: ∆time heightlocal/velocityfall

Introduction State of the art Our method Results and Conclusions

Real-time - CPU - Local space movement

Height correction

Introduction State of the art Our method Results and Conclusions

Real-time - CPU - Packet handling

Introduction State of the art Our method Results and Conclusions

Real-time - GPU - Vertex shader

Placement Particle state

Introduction State of the art Our method Results and Conclusions

Real-time - GPU - Geometry shader

Billboard expansion Drops: lighting parameters Splash: animation frame

Introduction State of the art Our method Results and Conclusions

Real-time - GPU - Fragment shader

Texture fetch Shadowing Shading

Introduction State of the art Our method Results and Conclusions

Test settings

City model (by www.Daz3D.com) 780K polygons 140 textures (color + alpha mask + bump map) Rain 400 x 400 meters, 230 meters high 375M particles 3606 x 2880 and 32 x 2880 mipmapped atlases (x10) 21 animation frames of 48 x 32 Computer Intel R

Core

TM 2 Duo at 3 GHz

NVIDIA R

GeForce R GTX 280 with 1GB of memory

Screen size of 1280 x 720

Introduction State of the art Our method Results and Conclusions

Results

Introduction State of the art Our method Results and Conclusions

Results

Introduction State of the art Our method Results and Conclusions

Results

Introduction State of the art Our method Results and Conclusions

Performance

Local space analysis

Introduction State of the art Our method Results and Conclusions

Performance

Radius analysis

Introduction State of the art Our method Results and Conclusions

Performance

Packet size analysis

Introduction State of the art Our method Results and Conclusions

Performance

Light amount analysis

Introduction State of the art Our method Results and Conclusions

Conclusions

✓ Realistic raindrop highlights ✓ Lighting considering shadows ✓ Easy rain configuration:

rain space density map

✓ Scene interaction:

no indoor rain splashes

✓ Good performance:

Particle packets handled with few and fast operations Per-particle operations only in GPU

Introduction State of the art Our method Results and Conclusions

Limitations

✗ Drop blending needs a depth buffer

→ no semi-transparent geometry → no volumetric data

✗ Tighter bounds on simulation volume hindered by present

• rganization

✗ Huge impact on performance due to light sources ✗ Unrealistic and repetitive splashes

Introduction State of the art Our method Results and Conclusions

End

Thanks! Questions?