“ Interactive 3D Modeling with 
 a Generative Adversarial Network “ JERRY LIU, FISHER YU, THOMAS FUNKHOUSER, Princeton University CVPR 2017

ͲΜͳ΋ͷʁ • GANΛ࢖͍ɺॳ৺ऀͷ3DϞσϦϯάΛࢧԉ • ϘΫηϧάϦουΛ࢖͍ɺϢʔβʔ͕෺ମͷ
 ֓ܗΛೖྗɻͦΕʹର͠GAN͕ΑΓϦΞϧͳ
 3D෺ମΛఏҊ

ઌߦݚڀͱൺ΂ͯͲ͕͍͜͢͝ʁ • ΄ͱΜͲͷ3DϞσϦϯάπʔϧʢMayaͳͲʣ͸ɺઐ໳Ո޲͖ ʹઃܭ͞Ε͓ͯΓɺॳ৺ऀ͕ΧδϡΞϧʹ࢖͏͜ͱ͸೉͍͠ • ͦͷͨΊ3Dܗঢ়Λهड़͢ΔͨΊͷ؆୯ͳΠϯλʔϑΣʔε͕ ఏҊ͞Ε͖ͯͨɻ • ۂઢΛεέονɺδΣενϟΛهड़ɺཱํମΛூΔ ͳͲ • ݁ہܗΛܾΊΔͷ͸ϢʔβʔͰ͋Δͱ͍͏఺Ͱݶք͕ଘࡏ • ຊ࿦จͷख๏Ͱ͸ɺϢʔβʔ͕֓ܗΛ͚ࣔͩ͢ͰɺࣗಈͰ
 ͦΕʹࣅͨ෺ମͷৄࡉͳܗঢ়ΛఏҊ͢Δ͜ͱ͕Ͱ͖Δ

ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ ᶃ ϢʔβʔͷϘΫηϧσʔλ x ͔Βજࡏۭؒ΁ࣹӨ͢Δ
 ωοτϫʔΫ P Λֶश ᶄ P ʹΑͬͯੜ੒͞Εͨ z Λ͔ͭͬͯੜ੒ ੜ੒ͷϑϩʔ G(z) P ( x ) z x x 0

ٕज़΍ख๏ͷΩϞ͸Ͳ͜ʁ • G, D ͷωοτϫʔΫߏ଄͸ [Wu et al. 2016] ͱ΄ͱΜͲಉ͡ min G max D log D(x) + log(1 D(G(z))) • P ͸ D ͱࣅͨߏ଄ͱͳ͍ͬͯΔɻҎԼͷΑ͏ʹֶशɻ P ( x ) = argmin z E ( x, G ( P ( x ))) E ( x, x 0) = 1D ( x, x 0) + 2R ( x 0)

Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ

Ͳ͏΍ͬͯ༗ޮͩͱݕূͨ͠ʁ • ࣦഊྫ • ࣅ͍ͯͳ͍෺ମ͕ੜ੒͞ΕΔ • ܗঢ়่͕ΕΔ • ΑΓચ࿅͞ΕͨόϦσʔγϣϯ
 ΍ޙॲཧΛ༻͍ͯվળͰ͖Δ • Ϣʔβʔͷೖྗͱֶशͨ͠෺ମͱͷτϨʔυΦϑͷόϥϯε • ΑΓߴ඼࣭ͳੜ੒͕Ͱ͖ΔωοτϫʔΫͷ໛ࡧ