PARVAI - HVS Aware Adaptive Display Power Management for Mobile Games

+

PARVAI - HVS Aware Adaptive Display Power

Management for Mobile Games

Bhojan Anand, Li Kecen, Akkihebbal L. Ananda

+Limitations of Mobile Computing

Weight Size

Performance Energy

+Well Known Problem

Demand and Supply

+Where does the energy go?

Bhojan Anand et al., ‘Adaptive Display Power Management for Mobile Games’, Mobisys

2011

HOW DID WE MEASURE?

+Why Organic LED (OLED)?

Alternate display technology to LCD. (Growing faster!)

Samsung Panasonic Motorola Nokia HTC LG

> 20 models 102p RAZR i Lumia820 One-S

Optimus

5.5” note II

Larger

Screens

7.7” Galaxy Tab

Much wider use – TVs,

Cameras, etc

55”

OL

ED

TV

+Why bother?

* S

ou

rces: E

ng

adget&

LG

Dis

pla

y

+Why Games?

Candy Crush Brings

In An Estimated

$633,000 A Day

© Alex Cocotas,

provided by

BUSINESS INSIDER

+Why Games?

Demand and Supply

HTC

Desire

HD

8h 10

min,

420h

5h 20

min, 490h

1h 50 min

+Power Consumption of Organic LED

(OLED)

Brighter contents

consumes more power

Each pixel is individually

illuminated in OLEDs

Factor 1: Brightness / Luminance

700 mW

255 mW

+Power Consumption of Organic LED

(OLED)

Disparity in OLED sub-pixel power

consumption

Factor 2: Color / Hue

two sub-pixels per pixel

RGBG - Pentile

Eg. Blue OLED material

Lowest luminance efficiency

(lu/Watt)

Higher current is used to

match luminance with Green

Results in Lower Lifetime

Solution: Larger Blue Sub-pixel.

(Variation in Size & Current for

each sub-pixel)

+Power Consumption of Organic LED

(OLED)

For any given color, we can generate power

efficient color with same brightness level

Factor 2: Color / Hue (Google Nexus One )

+Power Consumption of Organic LED

(OLED)

a) Nexus S

b) Galaxy S2

c) Galaxy Nexus

d) Galaxy S3

Dashed lines ->

normalized to 4

inch2 display.

+Basic Power Saving Approaches

Darken the contents

Make everything Green or Red

+HVS Unequal sensitivity to different colours

• The blue cones detect only colour due to low relative sensitivity.

• Light from the blue subpixels does little to help the eye resolve

images, most of it goes to waste.

+HVSSensitive to local contrast

+HVSArea of Interest

+PARVAI System

We leverage on HVS characteristics to

manipulate the contents with minimum visual

distortion

+Power Efficient Colour Mapping

For each color C = {r, g, b},

Map to a new color C’= {r’,g’,b’} with

- Minimum power consumption

- Minimum b’

With constraints:

- Hue change

- Saturation change

- Constant brightness/luminance

(Optimization problem)

Input: bLUT - (Look Up Table ordered by Blue) – All Colors (C)- 32bits

Output: bLUT with power efficient color C’

- b’ should be less than b

- Only Top half is searched

+Grid Based Transformations

Local colors and intensities are important

Pixel-by-pixel transformation is computation

intensive

Use Grid Based Transformations with opengl

Blend Function to Blend the colors to get the

target color

HVS Sensitive to local contrast

+Power Efficient Colour Mapping

Original

63 mW

Colour Mapped

55 mW

13% Power Saving

+Saliency Based Darkening

Center Part is more important then regions

Greater pressure when the sight angle is moving

vertically

For 640x480 screen, 480/640 = 0.75

For each gird cell X,Y:-

Saliency Based Darkening

Distance = abs(X+0.5-Xcenter) * 0.75 + abs(Y+0.5-Ycenter)

+Saliency Based Darkening

(187 mW) (151 mW)20% power saving

+Saliency Based Darkening

+Two Modes

Conservative:-

Quality Comparable to Original

Aggressive

Applies Power Efficient Color Mapping

Darker content outside the Area of Interest

Applies only Power Efficient Color Mapping

Power Efficient Color

Mapping

Saliency Based

Darkening+

+ Final Run-Time AlgorithmInput: Frame, Mode (aggressive or conservative)

Output: Power efficient Frame

Segment Frame to Virtual Grids

For Each Grid

Compute RAVE, GAVE, BAVE

Blue

Dominant?Color Map using bLUT

Aggressive

mode?

Calculate Distance and

Apply Darkening

+ EvaluationOriginal Conservative (Best Quality)

GCL: 0.75; MSSIM: 0.82 (6%)GCL: 0; MSSIM: 1

GCL: 4.488; MSSIM: 0.70 (27%)GCL: 3.713; MSSIM: 0.85 (27%)

Aggressive Simple Linear Darkening

+ Evaluation

Original Conservative (Best Quality)

Aggressive Simple Linear DarkeningGCL: 10.466; MSSIM: 0.78 (47%)

GCL: 1.411; MSSIM: 0.81 (13%)

GCL: 9.283; MSSIM: 0.80 (47%)

+Evaluation - User Study

Images and Game Play Video

Randomly selected images from a pool of

Images.

Randomly Selected Version: Original,

Simple Darkening, Conservative and

Aggressive

Select the Best - Colour, Brightness, Clarity

+Evaluation - Demography

+Evaluation – User Study (Image)

+Evaluation – User Study (Video)

+Contribution

Color Mapping Algorithm for Energy Efficiency

leveraging on the Non-linear response of HVS

System

Saliency based gradual darkening – uses non-

linear pressure in change of sight angle

Can save about 10% energy without any loss in

Visual quality - Conservative

Can save upto 45% energy with some losses in

Visual quality - Aggressive

Our approach

+Limitations & Future Work

User study with Game play

Evaluation with more quality metrics

Eg. Video Quality Evaluation Metrics

Generalisation to multiple devices

Application on Videos

Pre-computation

Applying on the Static Textures files

+Limitations & Future Work If the color fidelity is not very Important, then Power

Efficient Color mapping constraints can be relaxed further

Demo Videos: https://www.dropbox.com/sh/ez6huzfvks75h68/-XDbINocUt

+

Thank you!