gset somi a game specific eye tracking dataset for somi
play

GSET Somi: A Game-Specific Eye Tracking Dataset for Somi Hamed - PowerPoint PPT Presentation

GSET Somi: A Game-Specific Eye Tracking Dataset for Somi Hamed Ahmadi 1 Saman Zadtootaghaj 1 Sajad Mowlaei 1 Mahmoud Reza Hashemi 1 Shervin Shirmohammadi 1,2 University ofTehran 1 Multimedia Processing Laboratory (MPL), School of Electrical and


  1. GSET Somi: A Game-Specific Eye Tracking Dataset for Somi Hamed Ahmadi 1 Saman Zadtootaghaj 1 Sajad Mowlaei 1 Mahmoud Reza Hashemi 1 Shervin Shirmohammadi 1,2 University ofTehran 1 Multimedia Processing Laboratory (MPL), School of Electrical and Computer Engineering, College of Engineering, University of Tehran 2 DISCOVER Lab, School of Electrical Engineering and Computer Science, University of Ottawa

  2. G AMING I NDUSTRY • Wide range of gaming devices • Gaming will hit $91.5 billion this year 1 1 http://www.gamesindustry.biz/articles/2015-04-22-gaming-will-hit-usd91-5-billion-this-year-newzoo 7 th ACM Multimedia Systems Conference (ACM MMSys), 2 Klagenfurt am Wörthersee, Austria, May 2016

  3. C LOUD G AMING Bandwidth! Video Decoder Video Encoder Display System Game Engine Game Input Device Client Cloud 7 th ACM Multimedia Systems Conference (ACM MMSys), 3 Klagenfurt am Wörthersee, Austria, May 2016

  4. B ANDWIDTH C HALLENGE • Currently requires ~5Mbps per player • Perceptual video coding is used to reduce bit rate while preserving perceived quality! • Specific eye-tracking datasets are required to build specific perceptual models for gaming applications. 7 th ACM Multimedia Systems Conference (ACM MMSys), 4 Klagenfurt am Wörthersee, Austria, May 2016

  5. C OMPARISON OF THE G AME - RELATED E YE - TRACKING D ATASETS GSET P ETERS B ORJI CRCNS DIEM Collected ○ ○ ● ● ● while playing Collected ○ ○ ○ ● ● while watching Game video ○ ● ● ● ● Game video trailer ○ ○ ○ ○ ● #Subjects * 84 5 21 8 - #Videos * 135 24 27 - 4 Resolution 720p 680x480 680x480 680x480 Varying Video format Raw Raw MPEG-1 - H.264/AVC Eyes Both Right - - - Chin rest + Eye-tracker Remote Chin rest Chin rest - Head mount 7 th ACM Multimedia Systems Conference (ACM MMSys), 5 Klagenfurt am Wörthersee, Austria, May 2016

  6. V IDEO G AME • Title: “Somi, My Beautiful Doll” • Game Genre: Side-scrolling • Built by: GameMakerStudio • Resolution: 720p 7 th ACM Multimedia Systems Conference (ACM MMSys), 6 Klagenfurt am Wörthersee, Austria, May 2016

  7. SOMI’ S G AME O BJECTS • Categorized into eight groups 7 th ACM Multimedia Systems Conference (ACM MMSys), 7 Klagenfurt am Wörthersee, Austria, May 2016

  8. D ATA C OLLECTION P ROCEDURE Introduction Training Calibration Verification Playing 7 th ACM Multimedia Systems Conference (ACM MMSys), 8 Klagenfurt am Wörthersee, Austria, May 2016

  9. E YE - TRACKING D EVICE • Tobii X2-30 Compact ▫ Remote eye-tracker ▫ Sampling rate of 30 Hz ▫ Accuracy of 0.4 ° 7 th ACM Multimedia Systems Conference (ACM MMSys), 9 Klagenfurt am Wörthersee, Austria, May 2016

  10. S AMPLE R ESULTS 1 • Attention patterns are different among players of different skill levels Beginner Intermediate Expert 70 60 50 S KILL L EVEL S CORE R ANGE Percentage 40 30 Beginner score <= 1000 20 Intermediate 1000 < score <= 6000 10 Expert 6000 < score 0 Average attention per category 7 th ACM Multimedia Systems Conference (ACM MMSys), 10 Klagenfurt am Wörthersee, Austria, May 2016

  11. S AMPLE R ESULTS 2 • Attention patterns are different during different game states Beginner Intermediate Expert 70 Beginner Intermediate Expert 70 60 60 50 50 Percentage Percentage 40 40 30 30 20 20 10 10 0 0 Average attention per category in Running state Average attention per category in Jumping state 7 th ACM Multimedia Systems Conference (ACM MMSys), 11 Klagenfurt am Wörthersee, Austria, May 2016

  12. Dataset Structure • Each session contains ▫ Gaze records ▫ Keyboard strikes ▫ Mouse info ▫ Game objects’ info Size – Location – ▫ Gameplay video In a lossless format – http://www.site.uottawa.ca/~shervin/gaze/ 7 th ACM Multimedia Systems Conference (ACM MMSys), 12 Klagenfurt am Wörthersee, Austria, May 2016

  13. C ONCLUSION • Compared to existing datasets, ours has the following features at once: ▫ HD resolution ▫ Collection during gameplay instead of watching ▫ Recording of mouse and keyboard inputs ▫ Recording of game objects’ locations ▫ A large number of subjects • Can be used to recognize the different attention patterns among players 7 th ACM Multimedia Systems Conference (ACM MMSys), 13 Klagenfurt am Wörthersee, Austria, May 2016

  14. F UTURE W ORK • Adding more video games of the side-scrolling genre • Adding video games of the other game genres 7 th ACM Multimedia Systems Conference (ACM MMSys), 14 Klagenfurt am Wörthersee, Austria, May 2016

  15. 7 th ACM Multimedia Systems Conference (ACM MMSys), 15 Klagenfurt am Wörthersee, Austria, May 2016

Recommend


More recommend