Upgrade to Pro — share decks privately, control downloads, hide ads and more …

SIGGRAPH ASIA 2016で聴いてきた面白い論文達の紹介

Keisuke OGAKI
January 12, 2017

SIGGRAPH ASIA 2016で聴いてきた面白い論文達の紹介

12月に行われたコンピュータグラフィックスの国際学会SIGGRAPH ASIA 2016 (https://sa2016.siggraph.org/) で発表を聴いた論文のうち面白かったものを5つ共有します

Keisuke OGAKI

January 12, 2017
Tweet

More Decks by Keisuke OGAKI

Other Decks in Research

Transcript

  1. SIGGRAPH ASIA ใࠂձ
    ~ݸਓతਪ͠࿦จ BEST 5~
    େ֞ܚհ
    2016.12/21

    View Slide

  2. 1. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    2. EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    3. Gesture3D: Posing 3D Characters via Gesture Drawings
    4. Emptying, Refurnishing, and Relighting Indoor Spaces
    5. Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images

    View Slide

  3. 1. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    2. EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    3. Gesture3D: Posing 3D Characters via Gesture Drawings
    4. Emptying, Refurnishing, and Relighting Indoor Spaces
    5. Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images

    View Slide

  4. ΍ͬͨ͜ͱ
    ΫϩευϝΠϯͷσʔλΛಉۭؒ͡ʹϚοϐϯά͍ͨ͠
    1. Word2VecͳͲͰɺϥϕϧ৘ใͷۭؒΛ࡞Δ
    2. (ը૾༻CNNΛద౰ͳ໰୊Ͱࣄલֶश)
    3. 1ͱ2ͷۭؒΛ͚ۙͮΔΑ͏ʹ2Λֶश
    1. ͨͩ͠ɺ3DϞσϧʹ͍ͭͯ͸ϚϧνϏϡʔ
    CNN(ฏۉ)
    ͍͍ͱ͜
    • ڻ͖ͷগͳ͍࣮૷
    • ͍ΖΜͳυϝΠϯʹద༻ͯ͠Δ
    ѱ͍ͱ͜
    • ڻ͖ͷগͳ͍࣮૷
    Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images
    ϫγϯτϯେֶ 'BDFCPPL
    ࣗવݴޠΛ௒͑Δ௚ײతͳݕࡧख๏Λ࡞Ζ͏!

    View Slide

  5. 1. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    2. EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    3. Gesture3D: Posing 3D Characters via Gesture Drawings
    4. Emptying, Refurnishing, and Relighting Indoor Spaces
    5. Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images

    View Slide

  6. ΍ͬͨ͜ͱ
    1. ը૾܈͔Β෦԰Λ3DϞσϦϯά
    2. น΍চͳͲͷߏ଄ೝࣝ
    1. ௚ํମͷू·Γͱͯ͠ೝࣝͰ͖Δ΋ͷΛ෦԰
    ͱ͢Δ
    3. ޫݯҐஔਪఆ
    1. Image based lighting
    4. 3DϞσϧͷ࠶഑ஔ & ϦϥΠτ♪
    ͍͍ͱ͜
    σϞ͕Ϋʔϧ
    ѱ͍ͱ͜
    ٕज़త৽نੑʁ
    Emptying, Refurnishing, and Relighting Indoor Spaces
    IUUQTXXXZPVUVCFDPNXBUDI W4;ND64.
    ϫγϯτϯେֶ 'BDFCPPL
    ෦԰ͷࣸਅ܈͔ΒɺՈ۩Λফͯ͠ɺ࠶഑ஔ&ϦϥΠτ

    View Slide

  7. 1. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    2. EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    3. Gesture3D: Posing 3D Characters via Gesture Drawings
    4. Emptying, Refurnishing, and Relighting Indoor Spaces
    5. Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images

    View Slide

  8. ΍ͬͨ͜ͱ
    ΫϩευϝΠϯͷσʔλΛಉۭؒ͡ʹϚοϐϯά͍ͨ͠
    ͍͍ͱ͜
    • ເ͕͋Δ໰୊ઃఆ
    ѱ͍ͱ͜
    • ڪΒ͘ɺεέονࣗମͷਫ਼౓͕͔ͳΓߴ͍͜ͱΛٻΊΒ
    Ε͍ͯΔ
    Gesture3D: Posing 3D Characters via Gesture Drawings

    View Slide

  9. Gesture3D: Posing 3D Characters via Gesture Drawings
    εέονΛ෼ׂ͢Δ৚݅
    • ۂ཰͕มԽ͢Δ
    • ۂઢํ޲͕มԽ͢Δ
    ͳͲ
    • Ϙʔϯͷ੾Ε໨ͱεέονͷ੾Ε໨ͷڑ཭
    • Ϙʔϯͷ௕͞ͱεέονͷ͖ΕΊͷͳ͕͞
    • ্͔̎ͭΒಋ͔Ε֤ͨϘʔϯ͕ɺδϣΠ
    ϯτʹໃ६͠ͳ͍͔ͷ੍໿
    ෼ׂ͞Ε͚ͨͬͪ͢ͱϞσϧͷϚονϯά

    View Slide

  10. 1. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    2. EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    3. Gesture3D: Posing 3D Characters via Gesture Drawings
    4. Emptying, Refurnishing, and Relighting Indoor Spaces
    5. Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images

    View Slide

  11. ΍ͬͨ͜ͱ
    ͍͍ͱ͜
    • ઌߦݚڀ͔ΒͷࣗવͳൃలʹΈ͑Δ
    • ͚Ͳ࣮͸େ෼৽نੑ΋͋Δ
    ѱ͍ͱ͜
    • ઌߦݚڀ΋ͦ͏͚ͩͲɺجૅٕज़૊Έ߹Θͤ
    ·ͨ͠ɺͰ͸͋Δ
    EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    IUUQHWWNQJJOGNQHEFQSPKFDUT&HP$BQ
    ͨͬͨ̎ͭͷڕ؟ϨϯζͷΈΛ༻͍ͯϞʔγϣϯΩϟϓνϟʂʂʂ

    View Slide

  12. ઌߦݚڀ
    ਎ମதͷɺΩϟϓνϟ͍ͨ͠Ϙʔϯʹ͚ͭ
    ͨGoproΧϝϥΛ͔ͭͬͯϞʔγϣϯΩϟ
    ϓνϟ
    1. ਎ମதͷΧϝϥ͔ΒStructure from
    MotionͰɺ֎ଆͷ3DϚοϓͱΧϝϥҐஔ
    Λಉ࣌ਪఆ
    2. 1ͰಘΒΕͨΧϝϥҐஔͷ࣌ܥྻͱͯ͠
    ϞʔγϣϯΩϟϓνϟ
    EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras

    View Slide

  13. ৽نੑ
    ఏҊख๏Ͱ͸ɺΧϝϥ͸ͨͬͨೋ͔ͭ͠ଘ
    ࡏ͠ͳ͍ͷͰɺΧϝϥҐஔ = ؔઅͱ͍͏ͷ
    ͸Ͱ͖ͳ͍
    ͨͩ͠ɺڕ؟ϨϯζͳͷͰ಺ଆͷֆ΋औΕ
    Δ!!
    1. SfMͰ3DϚοϓΛ࡞Δ
    2. ڕ؟Χϝϥ͍ͨ͡ͷҐஔΛਪఆ͢Δ
    3. ڕ؟Χϝϥ͔Βࣗ෼ͷ֤ύʔπͷ૬ରҐ
    ஔΛೝࣝ͢Δ
    4. 2ͱ3͔Βۭؒ಺Ͱͷࣗ෼ͷಈ͖Λਪఆ͢
    Δ
    ͦΕͧΕʹڕ؟ͳΒͰ͸ͷ޻෉΋͋Δ
    EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras

    View Slide

  14. 1. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    2. EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras
    3. Gesture3D: Posing 3D Characters via Gesture Drawings
    4. Emptying, Refurnishing, and Relighting Indoor Spaces
    5. Shape2Vec: semantic-based descriptors for 3D shapes, sketches and images

    View Slide

  15. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    ͍͍ͱ͜
    • ݁Ռ͕͍͢͝
    • ख๏΋ଥ౰
    ѱ͍ͱ͜
    • ಉ͡Α͏ͳݚڀͯͨ͠ͷͰɺୡ੒
    ͞Εͯ͠·ͬͯେ͕֞൵͍͠
    700จࣈͷखॻ͖׽ࣈ͔Βɺ20000จࣈੜ੒

    View Slide

  16. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    (ετϩʔΫͷ࠲ඪ, 

    ετϩʔΫͷ֤఺ͷ࠲ඪ)
    -> (ετϩʔΫͷҐஔͷࠩ෼,
    ετϩʔΫͷ֤఺ͷҐஔ
    ͷࠩ෼)
    ͱ͍͏ֶश

    View Slide

  17. Automatic Generation of Large-scale Handwriting Fonts via Style Learning
    ը૾͔ΒͷετϩʔΫͷ੾Γग़
    ͠΋΍ͬͯΔ
    νϡʔϦϯάςετਖ਼౴཰50%
    ͗͢͢͝ʂʂʂʂ

    View Slide