PPTX - nossdav 2013

Mahdi Hemmati, Abbas Javadtalab,
Ali A. Nazari, Shervin Shirmohammadi, Tarik Arici
 Background
 GaV: Game as Video
 Advantages
 Challenges
 Proposed Method
 Evaluation
 Results
 Conclusion
 Future Work
 Cloud Gaming: real-time game playing via thin clients
(Cloud Computing + Online Gaming)
 Great interest and growth during recent years
 Several cloud gaming services with a variety of
realizations available on the market:
Background – cont.
 Cloud hosts for the game logic and streams the game
experience to the client
 Game Streaming:
 Streaming the 3D Objects
 Streaming the Rendered Video
 Hybrid Approach
(Classical approach)
(OnLive, GaiKai)
 Our Focus: Video Streaming
“Game as Video” (GaV)
 A natural combination:
 Cloud gaming + Mobile gaming (i.e., on mobile clients)
GaV Advantages
 No need for continuous hardware upgrade
 The only requirements are broadband internet connection
and a thin client (a device capable of video display)
 No need to purchase new versions of the games
 Pay as you play
 Play anywhere anytime
 Play the same game on various devices
(Smartphone, Tablet, Notebook, Desktop PC, Smart TV)
 Revenue increase for developers/Publishers by
leaving out the retail chain
GaV Challenges
 Stringent requirements of network service quality
 Network Bandwidth
GaV streaming data rates are significantly higher than
conventional gaming and similar to video streaming
 Latency
Network latency as well as available network bandwidth
greatly affects the player's quality of experience (QoE)
 Energy consumption of the servers in the Cloud
 Massive number of simultaneous game sesions
Proposed Method - Overview
 Basis: Our previous successful experience with activity-
based object selection for 3D object streaming
 Difference: rendering and video encoding done on server
side and only the encoded video streamed to the client
 Objective: adapt the game scene to achieve
 Lower video bit rate
 Faster encoding time at server side
(Lower energy consumption)
 Key Idea: exclude less important objects from the game
scene before rendering and encoding
Proposed Method:
Activity-based Object Selection
 Maintaining a list containing the importance of each
object for each activity, designed by game designers
Evaluating the importance of each object in each frame of
the game based on the current activity of the player
Optimizing object selection using their normalized
importance factors subject to some constraints
Rendering the scene containing only the selected objects
Encoding and streaming the video of the gameplay
Evaluation - Game
 Object selection algorithm implemented
in Unity 3D game engine
 Two Unity 3D Demo Games
 Video of the game play captured by FRAPS
Evaluation - Video
 Capture video encoded using x264 (H.264/AVC)
 Profile: High
 Rate control methods: ABR & CRF
 Target bit rate: 1Mbps
 Encoding time recorded using Intel VTune Amplifier
 Performance Metrics
 Size of the coded videos
 Streaming bit rates
 Encoding Time
BootCamp Game Screenshots
Results for the BootCamp Game
AngryBots Game Screenshots
Results for the AngryBots Game
Summary & Conclusion
 A game scene adaptation using an object selection and
optimization method proposed for GaV scenario
Only the most important objects from the perspective
of the player’s activity are encoded in the scene and
irrelevant or less important objects are omitted
Significantly lower streaming bit rates achieved
(between 2.2% to 8.8% less than the original video)
Slightly less processing time on server side
(still critical due to massive number of game sessions)
A complementary approach to existing methods, such
as low-polygonal modeling and level of detail scaling
Future Work
 Subjective evaluation of the quality of experience (QoE)
 Our previous work for client-side rendering
 Comparison of QoE: proposed scheme vs. the strategy of
higher compression of the entire scene with all objects
 Rendering less-important objects with a lower LoD
 Encoding less-important regions with lower bit rates
 Energy-aware video encoding algorithms

similar documents