game dev light - Speaker Deck

Slide 1

Slide 1 text

͜Ε͔ΒͷϏσΦήʔϜ࢈ۀʹ͍ͭͯ ը૾ͳ͠൛

Slide 2

Slide 2 text

೥ޙΛݟਾ͑Δ ͜Ε͔ΒͷϏσΦήʔϜ࢈ۀʹ͍ͭͯ

Slide 3

Slide 3 text

IUUQTXXXVOSFBMFOHJOFDPNKBCMPHXFMDPNFUPVOSFBMFOHJOF GGXΛඪ४γΣʔμͱ͢ΔUnreal Engine 4͕ϦϦʔε͞ΕΔ 2014೥ Unity͕ඪ४γΣʔμΛGGXʹมߋ IUUQTVOJUZEDPNKQVOJUZXIBUTOFXVOJUZ 2015೥

Slide 4

Slide 4 text

IUUQTXXXVOSFBMFOHJOFDPNKBCMPHXFMDPNFUPVOSFBMFOHJOF GGXΛඪ४γΣʔμͱ͢ΔUnreal Engine 4͕ϦϦʔε͞ΕΔ ࣍ʹϏσΦήʔϜ࢈ۀʹ޿·Δٕज़͸ಡΊΔ GGXʹؔ͢Δ࠷ॳͷ࿦จ Bruce Walter, Stephen R. Marschner, Hongsong Li, and Kenneth E. Torrance. 2007. Microfacet models for refraction through rough surfaces. In Proceedings of the 18th Eurographics conference on Rendering Techniques (EGSR'07), Jan Kautz and Sumanta Pattanaik (Eds.). Eurographics Association, Aire-la-Ville, Switzerland, Switzerland, 195-206. DOI=http:// dx.doi.org/10.2312/EGWR/EGSR07/195-206 2007೥ Stephen McAuley, Stephen Hill, Naty Hoffman, Yoshiharu Gotanda, Brian Smits, Brent Burley, and Adam Martinez. 2012. Practical physically-based shading in film and game production. In ACM SIGGRAPH 2012 Courses (SIGGRAPH '12). ACM, New York, NY, USA, Article 10, 7 pages. DOI: https://doi.org/10.1145/2343483.2343493 σΟεχʔ͕GGXΛ࢖͍ͬͯΔࣄΛެද 2012೥ 2014೥ Unity͕ඪ४γΣʔμΛGGXʹมߋ IUUQTVOJUZEDPNKQVOJUZXIBUTOFXVOJUZ 2015೥

Slide 5

Slide 5 text

ࠓݚڀ͕੝ΜʹߦΘΕ͍ͯͯ ਺೥Ҏ಺ʹήʔϜ࢈ۀͰԠ༻͕޿͕Γͦ͏ͳٕज़͸ ػցֶश Machine Learning ಛʹήʔϜ԰ͷ࢓ࣄΛେ͖͘ม͑ͦ͏ͳͷ͕ ఢରతੜ੒ωοτϫʔΫͱਂ૚2ωοτϫʔΫ

Slide 6

Slide 6 text

https://www.reddit.com/r/zelda/comments/5olf6h/ oc_i_fixed_my_zelda_map_size_comparison_graphic/ θϧμͷ఻આ࣌ͷΦΧϦφ θϧμͷ఻આ෩ͷλΫτ θϧμͷ఻આ τϫΠϥΠτϓϦϯηε 5IF&MEFS4DSPMMT7 4LZSJN θϧμͷ఻આ ϒϨεΦϒβϫΠϧυ

Slide 7

Slide 7 text

https://www.reddit.com/r/gaming/comments/5p6db2/ zelda_breath_of_the_wild_fair_map_comparison/ 5IF8JUDIFS8JME)VOU (SBOE5IFGU"VUP7 5IF&MEFS4DSPMMT7 4LZSJN θϧμͷ఻આ ϒϨεΦϒβϫΠϧυ

Slide 8

Slide 8 text

ͭͷήʔϜΛ࡞ΔͨΊʹ ࡞Βͳ͚Ε͹ͳΒͳ͍෺͸૿͍͑ͯΔ ͭͷήʔϜΛ࡞ΔͨΊʹ ࢖͏͜ͱ͕Ͱ͖ΔϦιʔε͸༗ݶͰ͋Δ ͍͔ʹޮ཰Α͘ήʔϜΛ࡞Δ͔͕ ॏཁʹͳ͍ͬͯΔ

Slide 9

Slide 9 text

Coins $ - 7 gold pieces ---------- ------------ Armor #-........| ##...........+####### ---------- a - an uncursed +2 pair of leather gloves (being worn) |.........#####|.....)....| #####|........| b - an uncursed +1 robe (being worn) .........| ####...........-# #|........| Comestibles |........| # ------------# #|<.......| g - 3 uncursed food rations |........| # # #-........- h - 6 uncursed apples ----------## # #|........| i - 8 uncursed oranges #### # #---------- j - 4 uncursed fortune cookies # # ########### l - a food ration # # # # m - 2 jackal corpses ####### # # p - a kobold corpse ## ## ## Scrolls -------# # # d - an uncursed scroll of identify |[email protected]# # ## n - a scroll labeled ZLORFIK |...>.|# #------------# Spellbooks |.....| #|..........|# c - a blessed spellbook of protection |...... #...........-# Potions |.....| |..........| e - 2 uncursed potions of healing ------- --------|--- f - a blessed potion of healing # Tools Foo the Candidate St:15 Dx:14 Co:12 In:9 Wi:1 k - a magic marker (0:43) Dlvl:1 $:7 HP:14(14) Pw:5(5) AC:4 Xp:1 Stressed Gems/Stones o - 6 rocks (end) େ͖ͳ෺Λখ͘͞࡞Δʹ͸ ࣗಈੜ੒ ೥୅ॳ಄ʹ͸طʹ μϯδϣϯΛࣗಈੜ੒͢ΔήʔϜ͕࡞ΒΕ͍ͯͨ

Slide 10

Slide 10 text

ۙ୅తͳ%ήʔϜͰ ࣗಈੜ੒Λߦ͓͏ͱ͢ΔͱͲ͏ͳΔ͔

Slide 11

Slide 11 text

https://www.gdcvault.com/play/1020197/Landscape-Creation-and-Rendering-in ۙ୅తͳ%ήʔϜͰ ࣗಈੜ੒Λߦ͓͏ͱ͢ΔͱͲ͏ͳΔ͔ ஍ܗΛਓ͕ؒ༻ҙͯ͠ ຊ෺Β͍͠ςΫενϟͱ ຊ෺Β͍͠৭ͱ഑ஔͷ૲໦Λੜ੒͢Δ ͜Ε͚ͩͰ΋ ෳࡶͳϧʔϧ Λཁ͢Δ

Slide 12

Slide 12 text

https://minecraft.net/ja-jp/ ࢥ͍੾ͬͯࣗಈੜ੒͠΍͘͢ͳΔ੍໿ͷ΋ͱͰ஍ܗΛ࡞Δ ݡ͍બ୒Ͱ͸͋Δ͕ ൚༻తʹ࢖͑Δํ๏Ͱ͸ͳ͍ ޿େͳੈքΛ খ͘͞࡞Δ͜ͱʹ ੒ޭ͍ͯ͠Δ

Slide 13

Slide 13 text

େ͖ͳ෺Λখ͘͞࡞Δʹ͸ ࣗಈੜ੒ ͕༗ޮ͕ͩ ΠϚυΩͷήʔϜͷίϯςϯπΛ ࣗಈੜ੒͢ΔϧʔϧΛߟ͑Δͷ͸༰қͰ͸ͳ͍

Slide 14

Slide 14 text

? ԿΒ͔ͷϧʔϧ ͜͏͍͏࣌ ͜͏ͳͬͯཉ͍͠ ࣗಈੜ੒͸ԿͰ΋ϥϯμϜʹੜ੒͢Ε͹ྑ͍Θ͚Ͱ͸ͳ͘ ଟ͘ͷ৔߹͜Μͳ෩ʹੜ੒͍ͨ͠ͱ͍͏໨ඪ͕͋Δ

Slide 15

Slide 15 text

? ԿΒ͔ͷϧʔϧ ͜͏͍͏࣌ ͜͏ͳͬͯཉ͍͠ ط஌ ط஌ ະ஌ ͍͔ͭ͘ͷࣄྫ͔Β ͦͷഎޙʹ͋Δϧʔϧ͕ԿͰ͋Δ͔Λߟ͑Δ໰୊ʹͳΔ ࣗಈੜ੒͸ԿͰ΋ϥϯμϜʹੜ੒͢Ε͹ྑ͍Θ͚Ͱ͸ͳ͘ ଟ͘ͷ৔߹͜Μͳ෩ʹੜ੒͍ͨ͠ͱ͍͏໨ඪ͕͋Δ

Slide 16

Slide 16 text

ػցֶश ? ͜ΕΛਓ͕ؒߟ͑ΔͷͰ͸ͳ͘ ܭࢉػΛ࢖ͬͯٻΊΑ͏ͱ͍͏ͷ͕ ೖྗ ग़ྗ ͍͔ͭ͘ͷࣄྫ͔Β ͦͷഎޙʹ͋Δϧʔϧ͕ԿͰ͋Δ͔Λߟ͑Δ໰୊ʹͳΔ Machine Learning

Slide 17

Slide 17 text

ڱ͍ ޿͍ ఢ͕ଟ͍ ఢ͕গͳ͍ ෦԰ͷಛ௃ ͷ͓खຊ͸͜͜ͰๅശΛग़͢΂͖ͩͱݴ͍ͬͯΔ ͷ͓खຊ͸͜͜ͰๅശΛग़͢ͷ͸ૣ͍ͱݴ͍ͬͯΔ ͷ৚݅Ͱ࡞ΒΕͨ෦԰ʹๅശ͸ग़Δ΂͖͔ ๅശग़ͤ೿ ๅശ·ͩ೿

Slide 18

Slide 18 text

ڱ͍ ޿͍ ఢ͕ଟ͍ ఢ͕গͳ͍ ෦԰ͷಛ௃ ͳΜͱͳ͘͜ͷลͰ෼͔Εͯͦ͏͔ͩΒๅശΛग़ͦ͏ ͷ৚݅Ͱ࡞ΒΕͨ෦԰ʹๅശ͸ग़Δ΂͖͔ ๅശग़ͤ೿ ๅശ·ͩ೿

Slide 19

Slide 19 text

ڱ͍ ޿͍ ఢ͕ଟ͍ ఢ͕গͳ͍ ෦԰ͷಛ௃ αϙʔτϕΫλϚγϯ ૬൓͢Δछྨͷ͓खຊ͔Β࠷େͷڑ཭ΛऔΕΔͱ͜ΖͰઢҾ͖ Support Vector Machine

Slide 20

Slide 20 text

ڱ͍ ޿͍ ఢ͕ଟ͍ ఢ͕গͳ͍ ͬͪͩͬͨ͜Β ๅശ͕ग़Δ ͬͪͩͬͨ͜Β ๅശ͕ग़ͳ͍ ෦԰ͷಛ௃ σʔλ͔Βϧʔϧ͕Ͱ͖ͨ

Slide 21

Slide 21 text

ڱ͍ ޿͍ ఢ͕ଟ͍ ఢ͕গͳ͍ ෦԰ͷಛ௃ ৚͕݅ෳࡶʹͳΔͱ ͜ͷํ๏ͰϧʔϧΛੜΈग़͢ͷ͸ແཧͦ͏

Slide 22

Slide 22 text

ڱ͍ ޿͍ ఢ͕ଟ͍ ఢ͕গͳ͍ ෦԰ͷಛ௃ ͲΜͳઢͰ෼ྨͰ͖Δ͔΋෼͔Βͳ͍ॴ͔Β ϧʔϧΛಘ͍ͨ

Slide 23

Slide 23 text

৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ ຊ෺Β͍͠஍ܗͷੜ੒ Éric Guérin, Julie Digne, Éric Galin, Adrien Peytavie, Christian Wolf, Bedrich Benes, and Benoît Martinez. 2017. Interactive example-based terrain authoring with conditional generative adversarial networks. ACM Trans. Graph. 36, 6, Article 228 (November 2017), 13 pages. DOI: https://doi.org/10.1145/3130800.3130804 ෼Ͱඳ͔Εͨεέον͔Βੜ੒͞Εͨ஍ܗ

Slide 24

Slide 24 text

೔ຊޠͰཔΉ ৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ ຊ෺Β͍͠஍ܗͷੜ੒

Slide 25

Slide 25 text

χϡʔϥϧωοτϫʔΫ ೖྗ ग़ྗ Neural Network ৚݅෇͖ఢରతੜ੒ωοτϫʔΫΛ༻͍ͨ ຊ෺Β͍͠஍ܗͷੜ੒ ૚ ܗࣜχϡʔϩϯ

Slide 26

Slide 26 text

ܗࣜχϡʔϩϯ ೖྗ1 ೖྗ2 ೖྗ3 ೖྗ4 ೖྗ5 w1 w2 w3 w4 w5 શͯͷೖྗʹ͋ΔॏΈwΛ ͔͚ͨ΋ͷͷ࿨ΛٻΊͯ ͜ΕΛແ਺ʹܨ͛ͨ΋ͷ͸ ॏΈ࣍ୈͰ ೚ҙͷؔ਺ΛۙࣅͰ͖Δ ReLU ग़ྗ ׆ੑԽؔ਺(ReLU) ʹ௨ͨ͠஋Λग़ྗ Formal Neuron

Slide 27

Slide 27 text

ͱΓ͋͑ͣద౰ͳॏΈΛઃఆͨ͠ωοτϫʔΫʹ ೖྗΛྲྀ͢ͱ ೖྗ ཉ͔ͬͨ͠ग़ྗͱ͸͍ͩͿҧ͏෺͕ग़ͯ͘Δ ≠ ཉ͔ͬͨ͠ग़ྗ ग़ྗ

Slide 28

Slide 28 text

ग़ྗ͕ͲΕ͘Β͍ཉ͔ͬͨ͠෺ʹ͍͔ۙΛද͢஋Λఆٛ͢Δ ଛࣦؔ਺ ໨ඪ ݁ߏࣅͯΔ ͪΐͬͱࣅͯΔ ࣅͯͳ͍ Loss Function

Slide 29

Slide 29 text

w0k w1k w2k w3k w4k yk x0 x1 x2 x3 x4 yk = R (∑ n wnk xn) YͱX͕෼͔͍ͬͯͯZΛٻΊΔํ๏ ∂E ∂wjk = ∂E ∂xk R′ k (∑ n wnk xn) xj XͷมԽ͕ଛࣦؔ਺ʹ༩͑ΔӨڹXʹΑΔภඍ෼ R′ k͸3F-6ΛLʹ͍ͭͯඍ෼ͨ͠΋ͷ R͸׆ੑԽؔ਺3F-6

Slide 30

Slide 30 text

∂E ∂wij = ∑ k ∂E ∂xk R′ k (∑ n wnk xn) wjk R′ j (∑ n wnj xn) xi i j k ࠷ޙ͔Β૚໨ͷภඍ෼͸ ߹੒ؔ਺ͷඍ෼Λ࢖ͬͯٻΊΔ ઌ΄ͲٻΊͨ࠷ޙͷ૚ͷภඍ෼ͷ஋Λ࢖͑Δ ࿈࠯཯ Chain Rule

Slide 31

Slide 31 text

ଛࣦؔ਺ͷ஋ ग़ޱଆ͔Βॱ൪ʹ XΛม͑Δͱग़ྗ͕Ͳ͏มΘΔ͔ ΛٻΊ͍ͯ͘ ޡࠩٯ఻೻๏ Back Propagation

Slide 32

Slide 32 text

ଛࣦؔ਺ͷ஋Λখ͍ͨ͘͞͠ ଛࣦؔ਺͕ࢁͩͱ͢Δͱ Ұ൪௿͍ॴʹߦ͖͍ͨ ͨͩ͠ࢁશମͷܗ͸Θ͔Βͳ͍ Xʹ͍ͭͯͷภඍ෼͸ ࠓཱ͍ͬͯΔ৔ॴͷ ܏ࣼ ܏ࣼʹԊͬͯ ΑΓ௿͍ํ΁Ҡಈ͍͚ͯ͠͹ Ұ൪௿͍ͱ͜ΖʹͨͲΓண͚Δؾ͕͢Δ ֶश Training ภඍ෼ͷ݁ՌʹԊͬͯXΛগ͠ಈ͔ͤ͹ ໨ඪͱ͢Δؔ਺ʹগۙͮ͘͠ؾ͕͢Δ

Slide 33

Slide 33 text

ޡࠩٯ఻೻ͰXΛमਖ਼ ͜ΕΛେྔͷσʔλʹ͍ͭͯ܁Γฦ͢ͱ χϡʔϥϧωοτϫʔΫ͸ ೖྗʹରͯ͠ظ଴͢Δग़ྗ͕ಘΒΕΔؔ਺ʹ͍͍ۙͮͯ͘ ೖྗ ଛࣦؔ਺ ग़ྗ σʔλ͔Βؔ਺͕Ͱ͖ͨ

Slide 34

Slide 34 text

∂E ∂wij = ∑ k ∂E ∂xk R′ k (∑ n wnk xn) wjk R′ j (∑ n wnj xn) xi ஈḪΔຖʹ׆ੑԽؔ਺ͷඍ෼ֻ͕͔Δ ͜ͷ஋͕খ͍͞ͱ্ͷ૚·ͰXͷमਖ਼͕ಧ͔ͳ͍ ReLU ReLUͷඍ෼ ඍ෼ͯ͠΋஋͕খ͘͞ͳΒͳ͍ ׆ੑԽؔ਺3F-6ͷൃݟʹΑΓ ௕͍ωοτϫʔΫ͕࣮༻తʹͳͬͨ σΟʔϓϥʔχϯά Deep Learning

Slide 35

Slide 35 text

ը૾ͷΑ͏ͳҐஔؔ܎Λ࣋ͬͨσʔλͰ͸ ࿈ଓͨ͠ϐΫηϧ͕Ͳ͏ฒΜͰ͍Δ͔͸ ॏཁͳ৘ใͰ͋Δ ? ΋͔ͯ͠͠ͳ

Slide 36

Slide 36 text

w13 w23 w33 w43 w53 y3 x1 x2 x3 x4 x5 w93 ʜ ͷΑ͏ͳԕ͘ͷϐΫηϧͱͷ઀ଓ͸ ແବʹͳΔՄೳੑ͕ߴ͍ w93

Slide 37

Slide 37 text

ը૾ͷہॴతͳಛ௃͸ ը૾ʹϑΟϧλΛద༻͢ΔࣄͰऔΓग़ͤΔ × ྫΤοδݕग़

Slide 38

Slide 38 text

ϑΟϧλ ೖྗ ग़ྗ ֶश͢Δը૾ॲཧϑΟϧλ ৞ΈࠐΈ Convolution

Slide 39

Slide 39 text

͜Μͳ෩ʹܨ͍Ͱ χϡʔϩϯΛݮΒ͍ͯ͘͠ࣄͰ ը૾͔Βॏཁͳಛ௃͚ͩΛ औΓग़͢ࣄΛࢼΈΔ େྔͷೖྗ ݶΒΕͨग़ྗ ը૾ʹԿ͕ඳ͔Ε͍ͯΔ͔Λ ೝࣝ͢ΔΑ͏ͳ৔߹ʹΑ͘࢖͏ ৞ΈࠐΈ Convolution

Slide 40

Slide 40 text

ٯ৞ΈࠐΈ େྔͷೖྗ େྔͷग़ྗ Ұ౓ॏཁͳಛ௃͚ͩΛऔΓग़͠ ৄࡉͳը૾ͷ෮ݩΛࢼΈΔ ը૾͔Βผͷը૾Λ ੜ੒͢Δ৔߹ʹΑ͘࢖͏ Deconvolution

Slide 41

Slide 41 text

৞ΈࠐΈ χϡʔϥϧωοτϫʔΫ Convolutional Neural Network

Slide 42

Slide 42 text

ਤʹ͢Δͷ͕ਏ͍ͷͰ Ҏ߱χϡʔϥϧωοτϫʔΫͷ૚Λ ͜ͷΑ͏ʹॻ͘

Slide 43

Slide 43 text

஍ܗͷੜ੒ ৞ΈࠐΈ ٯ৞ΈࠐΈ ϥϯμϜͳը૾Λྲྀ͢ͱ ຊ෺ͬΆ͍஍ܗͷߴ͞Ϛοϓʹͳͬͯ ग़͖ͯͯ΄͍͠

Slide 44

Slide 44 text

ଛࣦؔ਺ ஍ܗͷຊ෺ͬΆ͞ΛධՁ͢Δؔ਺͕ඞཁ Θ͔ΒΜ ຊ෺ͬΆ͍஍ܗ࡞Δ

Slide 45

Slide 45 text

෱Ԭ Ѫ஌ ळా ਓ޻Ӵ੕ʹΑͬͯଌྔ͞Ε࣮ͨࡍͷ஍ܗσʔλ͸ ൺֱత؆୯ʹखʹೖΕΔ͜ͱ͕Ͱ͖Δ

Slide 46

Slide 46 text

ຊ෺ͬΆ͍஍ܗ࡞Δ ຊ෺ͷ஍ܗ ੜ੒ͨ͠஍ܗ ஍ܗ͕ຊ෺͔൑ผ͢Δ΋͠൑ผʹࣦഊͨ͠ͳΒ ੜ੒ͨ͠஍ܗ͸ ຊ෺Β͍͠஍ܗͰ͋Δ ఢରతੜ੒ωοτϫʔΫ Generative Adversarial Network I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. C. Courville, and Y. Bengio. Generative adversarial nets. In Proceedings of NIPS, pages 2672– 2680, 2014. papers.nips.cc/paper/5423- generativeadversarial-nets.pdf.

Slide 47

Slide 47 text

ఢରతੜ੒ωοτϫʔΫ Generative Adversarial Network ຊ෺ͬΆ͍஍ܗ࡞Δ ஍ܗ͕ຊ෺͔൑ผ͢Δ ຊ෺ͷ஍ܗ ੜ੒ͨ͠஍ܗ ΋͠൑ผʹࣦഊͨ͠ͳΒ ੜ੒ͨ͠஍ܗ͸ ຊ෺Β͍͠஍ܗͰ͋Δ ্ԼͷωοτϫʔΫͷXΛަޓʹमਖ਼͢Δ ্ͷωοτϫʔΫͷXΛमਖ਼͢Δࡍʹ͸ԼͷωοτϫʔΫ͸ݻఆ ԼͷωοτϫʔΫ͕ ΑΓ൑ผͰ͖ͳ͍஍ܗ͕ੜ੒͞ΕΔΑ͏ʹमਖ਼Λߦ͏

Slide 48

Slide 48 text

ຊ෺ͬΆ͍஍ܗ࡞Δ ஍ܗ͕ຊ෺͔൑ผ͢Δ ຊ෺ͷ஍ܗ ੜ੒ͨ͠஍ܗ ΋͠൑ผʹࣦഊͨ͠ͳΒ ੜ੒ͨ͠஍ܗ͸ ຊ෺Β͍͠஍ܗͰ͋Δ ԼͷωοτϫʔΫͷXΛमਖ਼͢Δࡍʹ͸্ͷωοτϫʔΫ͸ݻఆ ຊ෺ͷ஍ܗͱ্ͷωοτϫʔΫ͕ੜ੒ͨ͠΋ͷΛ ΑΓਖ਼ࣝ͘͠ผͰ͖ΔΑ͏ʹमਖ਼Λߦ͏ ఢରతੜ੒ωοτϫʔΫ Generative Adversarial Network

Slide 49

Slide 49 text

ຊ෺ͬΆ͍஍ܗ࡞Δ ஍ܗ͕ຊ෺͔൑ผ͢Δ ຊ෺ͷ஍ܗ ΋͠൑ผʹࣦഊͨ͠ͳΒ ੜ੒ͨ͠஍ܗ͸ ຊ෺Β͍͠஍ܗͰ͋Δ ৚݅෇͖ఢରతੜ੒ωοτϫʔΫ Conditional Generative Adversarial Network εέονੜ੒

Slide 50

Slide 50 text

Slide 51

Slide 51 text

("/Λ࢖ͬͨ%00.ͷεςʔδͷੜ੒ (JBDPNFMMP &EPBSEPFUBMl%00.-FWFM(FOFSBUJPO6TJOH(FOFSBUJWF "EWFSTBSJBM/FUXPSLTz*&&&(BNFT &OUFSUBJONFOU .FEJB$POGFSFODF (&. ("/Ͱ ࣮ࡍͷ%ήʔϜͷεςʔδΛ ຊ෺Β͘͠ ੜ੒ͨ͠ࣄྫ ιʔείʔυ https://github.com/DanieleLoiacono/DoomGAN ࿦จ https://arxiv.org/abs/1804.09154

Slide 52

Slide 52 text

%00. ᴈ໌ظͷ'14 ιʔείʔυ͕ެ։͞Ε͍ͯͯվ଄͠΍͍͢ҝ ౓ʑݚڀऀͷ͓΋ͪΌʹͳΔ https://github.com/id-Software/DOOM

Slide 53

Slide 53 text

εςʔδΛ࡞Δ Ϛοϓ͕ຊ෺͔ࣝผ͢Δ ຊ෺ͷ %00.ͷεςʔδΛ ը૾ʹͨ͠΋ͷ ੜ੒ͨ͠εςʔδ ΋͠൑ผʹࣦഊͨ͠ͳΒ ຊ෺Β͍͠εςʔδͰ͋Δ ϝλσʔλ΍ ෦԰਺౳ͷ ಛ௃ྔ

Slide 54

Slide 54 text

࿦จͷஶऀʹΑΔσϞಈըIUUQTZPVUVCF,';UK21 ΍ͬͨࣄ͕͋ΔਓͰͳ͍ͱΘ͔Γʹ͍͔͘΋͠Εͳ͍͚Ͳ ࣮ࡍʹ͋Γͦ͏ͳܗͷεςʔδ͕ग़དྷ্͕Δ

Slide 55

Slide 55 text

$ZDMF("/ΛԠ༻ͨ͠υοτֆͷੜ੒ Chu Han, Qiang Wen, Shengfeng He, Qianshu Zhu, Yinjie Tan, Guoqiang Han, and Tien- Tsin Wong. 2018. Deep unsupervised pixelization. In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 243, 11 pages. DOI: https://doi.org/10.1145/3272127.3275082 http://www.shengfenghe.com/deep- unsupervised-pixelization.html ࣸਅ΍ΠϥετΛ ಥͬࠐΜͩΒ υοτֆʹͳͬͯग़ͯ͘Δ ͨͩॖখ͢ΔͷͰ͸ͳ͘ ಛ௃Λ͖ͬ͘Γ࢒ͯ͠ ϑϥοτͳ৭࢖͍ʹͳΔ ͜Ε͸ ਓ͕ඳ͍ͨ৔߹

Slide 56

Slide 56 text

A B B' ࣅ͍ͯΔ΄Ͳྑ͍ ը૾܈"ͱը૾܈#͕ ରͰରԠ͍ͯ͠Δ৔߹ "Λೖྗͱͯ͠ಘΒΕͨੜ੒෺ͱ#ͷҧ͍Λ ଛࣦؔ਺ʹֶͯ͠श͢Ε͹ྑ͍ "ͷΑ͏ͳը૾Λ#ͷΑ͏ͳը૾ʹม׵͍ͨ͠

Slide 57

Slide 57 text

$ZDMF("/ "Λ#ʹม׵ ຊ෺ͷ#ͱࠨͷੜ੒෺Λࣝผ #Λ"ʹม׵ ຊ෺ͷ"ͱӈͷੜ੒෺Λࣝผ A B ຊ෺ͷ#ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ ຊ෺ͷ"ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ #ˠ"ˠ#Ͱ ݩͱಉ͡ʹͳΔఔྑ͍ "ˠ#ˠ"Ͱ ݩͱಉ͡ʹͳΔఔྑ͍ Jun-Yan Zhu*, Taesung Park*, Phillip Isola, and Alexei A. Efros. "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks", in IEEE International Conference on Computer Vision (ICCV), 2017. (* indicates equal contributions)

Slide 58

Slide 58 text

$ZDMF("/ "Λ#ʹม׵ ຊ෺ͷ#ͱࠨͷੜ੒෺Λࣝผ #Λ"ʹม׵ ຊ෺ͷ"ͱӈͷੜ੒෺Λࣝผ A B #ˠ"ˠ#Ͱ ݩͱಉ͡ʹͳΔఔྑ͍ ຊ෺ͷ#ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ ຊ෺ͷ"ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ "ˠ#ˠ"Ͱ ݩͱಉ͡ʹͳΔఔྑ͍ ຊ෺ͷ"ͱຊ෺ͷ#͸ରͰରԠ͍ͯ͠Δඞཁ͸ͳ͍ "ͷྫͱ#ͷྫ͕୔ࢁ͋Ε͹૒ํ޲ͷม׵͕ಘΒΕΔ

Slide 59

Slide 59 text

υοτֆͷੜ੒ ֦େ ຊ෺ͷ"ͱӈͷੜ੒෺Λࣝผ A B ֦େˠॖখˠλονͰ ݩͱಉ͡ʹͳΔఔྑ͍ ॖখˠλονˠ֦େͰ ݩͱಉ͡ʹͳΔఔྑ͍ ࣸਅ΍Πϥετ υοτֆ ݩͷ৭ͱڥքΛ อ͍ͯͯΔఔྑ͍ ॖখ υοτֆͬΆ͍ λονʹ͢Δ ຊ෺ͷυοτֆͱੜ੒෺Λࣝผ͢Δ ຊ෺ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ ਅٯͷૢ࡞ʹ ͳ͍ͬͯΔఔྑ͍ http://www.shengfenghe.com/pixelization.html ͜͜ʹੜ੒ྫ͕୔ࢁ͋Δ

Slide 60

Slide 60 text

υοτֆͷੜ੒ http://www.shengfenghe.com/deep-unsupervised-pixelization.html

Slide 61

Slide 61 text

("/Λ୔ࢁ૊Έ߹Θͤͨ ౎ࢢͷσΟςΟʔϧͷੜ੒ Tom Kelly, Paul Guerrero, Anthony Steed, Peter Wonka, and Niloy J. Mitra. 2018. FrankenGAN: guided detail synthesis for building mass models using style-synchonized GANs. In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 216, 14 pages. DOI: https://doi.org/10.1145/3272127.3275065 ೖྗ ग़ྗ http://geometry.cs.ucl.ac.uk/projects/2018/frankengan/

Slide 62

Slide 62 text

೉ॴ ૭౳ͷ഑ஔͷੜ੒ɺৄࡉͳܗঢ়ͷੜ੒ɺ৭ͷੜ੒ ੜ੒͠ͳ͚Ε͹ͳΒͳ͍΋ͷ͕ଟ͍ શͯͷݐ෺͕ಉ͡ݟͨ໨ʹͳ͍ͬͯͨΒෆࣗવ ੜ੒݁Ռʹଟ༷ੑΛ࣋ͨͤΔඞཁ͕͋Δ

Slide 63

Slide 63 text

৚݅ cGAN ཚ਺ ৚݅ʹରԠ͢Δग़ྗ ֶशΛߦ͏ͨΊʹ͸ ৚݅ͱਖ਼͍͠ग़ྗͷηοτ͕ େྔʹඞཁ ৚݅෇͖ఢରతੜ੒ωοτϫʔΫͷ໰୊఺

Slide 64

Slide 64 text

GAN ཚ਺ ৚݅ແ͠ఢରతੜ੒ωοτϫʔΫͷ໰୊఺ ग़ྗΛগ͠ม͍͑ͨͱࢥͬͯ΋ ೖྗͷཚ਺ͷͲ͕͜มΘΕ͹ཉ͍͠ग़ྗʹͳΔͷ͔ શ͘෼͔Βͳ͍

Slide 65

Slide 65 text

ΦʔτΤϯίʔμ A A' ೖྗͱग़ྗ͕ ࣅ͍ͯΔఔྑ͍ ͜ͷ෦෼͕ࡉ͍ͷͰ׬શͳ৘ใΛग़ྗʹ఻͑Δࣄ͸ग़དྷͳ͍ ݶΒΕͨ৘ใ͔ΒݩͷσʔλΛ෮ݩ͢ΔͨΊʹ ॏཁͳಛ௃͸ͲΕ͔Λֶश͢Δ͜ͱʹͳΔ

Slide 66

Slide 66 text

ΦʔτΤϯίʔμ A A' ॏཁͳ ಛ௃ Τϯίʔμ σίʔμ

Slide 67

Slide 67 text

ཚ਺͔Β"Λੜ੒ A' A ຊ෺ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ Z ຊ෺ͷ"ͱੜ੒͞Εͨ"Λࣝผ ී௨ͷ("/

Slide 68

Slide 68 text

ཚ਺͔Β"Λੜ੒ A' A ຊ෺ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ Z ຊ෺ͷ"ͱੜ੒͞Εͨ"Λࣝผ *OGP("/ C ఢରతੜ੒ωοτϫʔΫ C' ࣅ͍ͯΔఔྑ͍ $͕"ͷॏཁͳಛ௃ʹؔ܎͍ͯ͠ΔͳΒ $ ; ˠ"ˠ$Ͱݩͷ$ʹ໭ΕΔഺ ΦʔτΤϯίʔμ $͕"ͷಛ௃Λίϯτϩʔϧ͢Δ ύϥϝʔλʹͳΔ "͔Β$Λ෮ݩ X. Chen, Y. Duan, R. Houthooft, J. Schulman, I. Sutskever, and P. Abbeel. Infogan: interpretable representation learning by information maximizing generative adversarial nets. In NIPS, 2016.

Slide 69

Slide 69 text

ཚ਺Λແࢹͯ͠Λੜ੒ ຊ෺ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ Z ΛࣝผͰ͖ͳ͍ Ϟʔυ่յ ࣝผث͕ ݟ෼͚ΒΕͳ͍෺ͳΒ ͳΜͰ΋ྑ͍ ೖྗͷཚ਺;Λແࢹͯ͠ ৗʹಉ͡Α͏ͳσʔλΛੜ੒͢Δ ωοτϫʔΫ͕ग़དྷ্͕Δࣄ͕͋Δ Mode Collapse

Slide 70

Slide 70 text

;͔Β"Λੜ੒ A' A ຊ෺ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ ຊ෺ͷ"ͱੜ੒͞Εͨ"Λࣝผ 7"&("/ ఢରతੜ੒ωοτϫʔΫ ΦʔτΤϯίʔμ ࣅ͍ͯΔఔྑ͍ ੜ੒ث͕;Λ࢖͍ͬͯΔͳΒ "ˠ;ˠ"Ͱݩͷ"ʹ໭ΕΔഺ Ϟʔυ่յΛى͜͢ͱ ΦʔτΤϯίʔμଆͷଛࣦ͕૿͑Δ "ͷॏཁͳಛ௃ΛؚΉ;Λ࡞Δ Rosca, Mihaela & Lakshminarayanan, Balaji & Warde-Farley, David & Mohamed, Shakir. (2017). Variational Approaches for Auto- Encoding Generative Adversarial Networks. Z

Slide 71

Slide 71 text

Z A' A ຊ෺ͱੜ੒෺Λ ࣝผͰ͖ͳ͍ఔྑ͍ #JDZDMF("/ C' C Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A. Efros, Oliver Wang: “Toward Multimodal Image-to-Image Translation”, 2017; arXiv:1711.11586 ࣅ͍ͯΔఔྑ͍ ࣅ͍ͯΔఔྑ͍ D("/Ͱ৚݅Λࢦఆ *OGP("/͕ ੜ੒෺ͷόϦΤʔγϣϯΛ࡞Γ 7"&("/͕Ϟʔυ่յΛ๷͙

Slide 72

Slide 72 text

A ֶश ৚݅ͱͦΕΛຬͨ͢"ͷϖΞΛ࢖ͬͯ Τϯίʔμͱੜ੒ثΛ࡞Δ C B ੜ੒ A "ͷॏཁͳಛ௃ͱ৚݅ͱଟ༷ੑͷૉ$͔Β "ͷಛ௃Λ࣋ͪ৚݅Λຬͨ͢ੜ੒෺Λ୔ࢁ࡞Δ #JDZDMF("/

Slide 73

Slide 73 text

্͔Βݟͨ ܗঢ় େࡶ೺ͳ ݐ෺ͷܗ ࣮ࡍͷ นͷը૾ ࣮ࡍͷ ԰ࠜͷը૾ ૭౳ͷ഑ஔͷநग़ Ԏಥ౳ͷ഑ஔͷநग़ ԣ͔Β ݟͨܗঢ় ԰ࠜͷԜತ ԰ࠜͷ໛༷ ૭ͷ഑ஔ นͷ໛༷ นͷԜತ ߴղ૾౓ͳ ԰ࠜͷ໛༷ ߴղ૾౓ͳ ૭ͷ໛༷ ૭࿮ͷܗঢ় ߴղ૾౓ͳ นͷ໛༷ ԰ࠜͷԜತ นͷԜತ ৄࡉͳ ݐ෺ͷ ܗͱ໛༷ FrankenGAN

Slide 74

Slide 74 text

χϡʔϥϧωοτϫʔΫ͕ ༏Εͨࣗಈੜ੒Λߦ͏ͨΊʹ͸ ଛࣦؔ਺ͷઃܭ͕伴ʹͳΔ ͦΕҎ֎͍͡ΕΔͱ͜Ζ͕ແ͍ͷͰ ग़͖ͯͯ΄͍͠෺Λ͍͔ʹͯ͠ଛࣦؔ਺Ͱදݱ͢Δ͔ʹͳΔ ͱ΋ݴ͑Δ

Slide 75

Slide 75 text

ڧԽֶश Reinforcement Learning ୡ੒͍ͨ͠໨ඪʹۙͮ͘ํ๏Λֶश͢Δ ঢ়گΛೖྗͱͯ࣍͠ͷߦಈΛग़ྗ͢Δ ͨͩͦ͠ͷߦಈ͕ਖ਼͔͔ͬͨ͠͸͙͢ʹ֬ఆ͢Δͱ͸ݶΒͳ͍ a = f(s) ࠓͷঢ়ଶ ࣍ʹ औΔ΂͖ߦಈ

Slide 76

Slide 76 text

ใु Reward Ձ஋ ӈ୺ͷϚεʹͨͲΓண͘ͱՁ஋ͷใु͕ಘΒΕΔͱ͢Δ

Slide 77

Slide 77 text

2ֶश Q Learning ͋ΔϚε͔ΒྡͷϚεʹҠಈͨ͠Β Ձ஋ͷ͓ๅ͕ஔ͍ͯ͋Δ࣌ ͦͷҠಈʹ͸ͷՁ஋͕͋Δ Ձ஋

Slide 78

Slide 78 text

2ֶश Q Learning Ձ஋ͷߦಈ͕Ͱ͖Δঢ়ଶʹ ભҠͰ͖Δߦಈʹ͸ ΑΓগ͠গͳ͍Ձ஋͕͋Δͱߟ͑Δ Ձ஋ Ձ஋

Slide 79

Slide 79 text

2ֶश Q Learning ͍Ζ͍Ζͳঢ়ଶ͔ΒߦಈͷՁ஋ΛٻΊ͍ͯ͘ Ձ஋ Ձ஋ ͜ͷૢ࡞Λे෼ͳճ਺܁Γฦ͢ͱ ֤ঢ়ଶͰߦ͑Δ֤ߦಈͷՁ஋͕໌Β͔ʹͳΔ

Slide 80

Slide 80 text

2ֶश Q Learning ͋Δঢ়ଶ͔Β͋ΔߦಈΛߦͳͬͨ৔߹ͷՁ஋ͷද ঢ়ଶ ߦಈ Ձ஋ ʜ ঢ়ଶ΍ߦಈͷ਺͕ଟ͍໰୊Ͱ͸ ͜ͷද͸େ͖͘ͳΓ͗ͯ͢هԱͰ͖ͳ͍

Slide 81

Slide 81 text

/FVSBM'JUUFE2*UFSBUJPO ͦͷߦಈʹ͸ ͜Ε͚ͩͷ Ձ஋͕͋Δ શͯͷߦಈͷՁ஋ΛهԱ͓ͯ͘͠୅ΘΓʹ ͋Δঢ়ଶ͔Β֤ߦಈΛߦͳͬͨ৔߹ͷՁ஋Λਪఆ͢Δ χϡʔϥϧωοτϫʔΫͷֶशΛߦ͏ ࠓ͜͏͍͏ঢ়ଶͰ ͜͏͍͏ߦಈΛ ͨ͠৔߹ Riedmiller, Martin. Neural fitted Q iteration–first experiences with a data efficient neural reinforcement learning method. In: European Conference on Machine Learning. Springer, Berlin, Heidelberg, 2005. p. 317-328.

Slide 82

Slide 82 text

ਪఆ ࣮ࡍ͸ ਪఆ ࣮ࡍ͸ ਪఆ ࣮ࡍ͸ ਪఆ ࣮ࡍ͸ Ձ஋ ʹḷΓண͚ͨ Ͱ΋ߦಈͷՁ஋͸ ͷ༧૝ͱҧͬͨ ࠓ"ͷঢ়ଶ ͜͏͍͏ଛࣦؔ਺ͰֶशΛߦ͏ ʹߦ͘ͱ͘Β͍ͷྑ͞ ͷྑ͞͸Ͳ͜ΖͰ͸ͳ͔ͬͨ ࣮ࡍͱͷ ޡࠩ "

Slide 83

Slide 83 text

ਪఆ ࣮ࡍ͸ ਪఆ ࣮ࡍ͸ ਪఆ ࣮ࡍ͸ Ձ஋ ʹḷΓண͚ͳ͍৔߹΋ಉ༷ Ձ஋ ʹߦ͘ͷ͕ߴՁ஋ ʹߦ͘ͷ͸ྑ͘ͳ͍ ࣮ࡍͱͷ ޡࠩ

Slide 84

Slide 84 text

Q(st , at ) ← Q(st , at ) + α (R(st , at ) + γmax (Q(st+1 , at+1 )) − Q(st , at )) XΛগ͠मਖ਼͢Δࡍͷগ͠۩߹ ࠓͷߦಈͰಘΒΕΔใु ଍ݩʹ͕͋Ε͹ͦͷՁ஋ मਖ਼લͷχϡʔϥϧωοτϫʔΫ͕ਪఆͨ͠ ࣍ͷߦಈΛߦͳͬͯભҠͨ͠ઌͷߦಈͷՁ஋ मਖ਼લͷχϡʔϥϧωοτϫʔΫ͕ਪఆͨ͠ ࠓͷߦಈͷՁ஋ ͜Εͱ͜ΕͷࠩΛখ͘͢͞ΔΑ͏ʹXΛमਖ਼͢Δ

Slide 85

Slide 85 text

Q(st , at ) ← Q(st , at ) + α (R(st , at ) + γmax (Q(st+1 , at+1 )) − Q(st , at )) ޡࠩΛܭࢉ͢ΔաఔʹχϡʔϥϧωοτϫʔΫͷग़ྗ͕ඞཁ ֶशத͸XΛαϯϓϧຖʹมߋ͍ͯ͠Δҝ͜ͷ஋͕҆ఆ͠ͳ͍ ਪఆͨ͠Ձ஋͕ ٻΊͨՁ஋ʹ͍ۙ΄Ͳྑ͍ ࣌ʑίϐʔ Xͷमਖ਼͕Ձ஋ͷਪఆʹ ௚͙ʹ൓ө͞Εͳ͍Α͏ʹͯ͠ ଛࣦؔ਺Λ҆ఆͤ͞Δ ਂ૚2ωοτϫʔΫ Deep Q Network Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013.

Slide 86

Slide 86 text

Vlad Firoiu, William F. Whitney, Joshua B. Tenenbaum Beating the World's Best at Super Smash Bros. with Deep Reinforcement Learning. arXiv preprint arXiv:1702.06230, 2017. ਂ૚QωοτϫʔΫΛ༻͍ͨ େཚಆεϚογϡϒϥβʔζDXͷAIͷ࡞੒ https://youtu.be/dXJUlqBsZtE ڧԽֶश ϑΝϧίϯ ϓϩήʔϚʔ ϑΝϧίϯ "*͕ϓϩͱઓͬͯউͬͨࣄΛใ͡Δهࣄ https://www.businessinsider.com/super-smash-bros- pro-slox-mafia-artificial-intelligence-2017-2

Slide 87

Slide 87 text

ਓؒ͸ը໘ΛݟͯߦಈΛܾఆ͢Δ "*͸ϝϞϦΛݟͯߦಈΛܾఆ͢Δ ͜ͷҧ͍͕͠͹͠͹ ΤϑΣΫτ͕දࣔ͞ΕΔલ͔Βߦಈͩ͢͠ͱ͍ͬͨ ཧෆਚʹڧ͍"*ΛੜΈग़͢

Slide 88

Slide 88 text

χϡʔϥϧωοτϫʔΫʹ ը໘͔Βঢ়ଶΛ൑அ͢Δͱ͜Ζ͔Β ΍ΒͤΔ͜ͱ͸ग़དྷͳ͍͔

Slide 89

Slide 89 text

Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013. Atari 2600ͷը໘Λೖྗͱͯ͠ ద੾ͳߦಈΛબ୒͢ΔڧԽֶश

Slide 90

Slide 90 text

Guillaume Lample, Devendra Singh Chaplot Playing FPS Games with Deep Reinforcement Learning. In: AAAI. 2017. p. 2140-2146. ը໘Λೖྗͱͯ͠ DOOMΛ߈ུ͢ΔڧԽֶश https://github.com/fhennecker/deepdoom https://youtu.be/L_9F9bO3-mI ࿦จͷஶऀʹΑΔσϞ ஶऀͱ͸ผͷਓʹΑΔ࣮૷

Slide 91

Slide 91 text

Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, Martin Riedmiller Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602, 2013. ը໘͔Β͙͢ʹ࣍͢΂͖ࣄ͕Θ͔ΔήʔϜͰ͸ ڧԽֶशͷείΞ͸ਓؒͷείΞΛ௒͑Δ ௕ظతͳઓུΛཁ͢ΔήʔϜͰ͸ਓ͕ؒѹউ͢Δ

Slide 92

Slide 92 text

ಛผͳ஍ܗʹԊͬͨಈ͖͸ ͦΕʹ߹ΘͤͯΩϟϓνϟΛߦͳ͓ͬͯ͘ඞཁ͕͋Δ ݹయతͳϞʔγϣϯΩϟϓνϟʹΑΔ Ξχϝʔγϣϯͷݶք

Slide 93

Slide 93 text

Daniel Holden, Taku Komura, and Jun Saito. 2017. Phase-functioned neural networks for character control. ACM Trans. Graph. 36, 4, Article 42 (July 2017), 13 pages. DOI: https:// doi.org/10.1145/3072959.3073663 http://theorangeduck.com/page/phase-functioned-neural-networks-character-control ϞʔγϣϯΩϟϓνϟͱ஍ܗͷϖΞͰ χϡʔϥϧωοτϫʔΫͷֶशΛߦ͍ ༩͑ΒΕͨ஍ܗΛࣗવʹา͘ಈ͖Λ ੜ੒͢Δ ஍ܗ ஍ܗʹԊͬͨ า͖ํ

Slide 94

Slide 94 text

࣍ʹલʹग़͢΂͖଍͕पظతʹʹมԽ͍ͯ͠Δ + + = = ಉ͡஍ܗͰ΋ λΠϛϯάʹΑͬͯ ద੾ͳಈ͖ํ͸มΘΔ w(0) w(1) w(2) w(3) पظతʹ੾ΓସΘΔͭͷXΛ࣋ͭωοτϫʔΫͰਪఆ 1IBTF'VODUJPOFE/FVSBM/FUXPSL

Slide 95

Slide 95 text

= = λΠϛϯάʹΑͬͯ ద੾ͳಈ͖ํ͸มΘΔ w(0) w(1) w(2) w(3) पظతʹ੾ΓସΘΔͭͷXΛ࣋ͭωοτϫʔΫͰਪఆ े෼ͳྔͷֶशσʔλΛूΊΔͨΊʹ ࣮ࡍʹ༷ʑͳ஍ܗΛ࣌ؒʹΘͨͬͯา͖ճͬͨ৔߹ͷಈ͖Λ ϞʔγϣϯΩϟϓνϟͰूΊΔ ΋͏ͪΐͬͱָͰ͖ͳ͍͔

Slide 96

Slide 96 text

ڧԽֶशʹΑΔาߦϞʔγϣϯͷੜ੒ Xue Bin Peng, Glen Berseth, Kangkang Yin, and Michiel Van De Panne. 2017. DeepLoco: dynamic locomotion skills using hierarchical deep reinforcement learning. ACM Trans. Graph. 36, 4, Article 41 (July 2017), 13 pages. DOI: https://doi.org/ 10.1145/3072959.3073602 https://www.cs.ubc.ca/~van/papers/2017-TOG-deepLoco/

Slide 97

Slide 97 text

ʮา͘ҝͷ଍ͷಈ͔͠ํ͕ະ஌ͷঢ়ଶ͔Β໨ඪ஍఺·Ͱา͚ͨΒใुʯ ͜Ε͸ใु͕ԕ্͗ͯ͢ख͍͔͘ͳ͍ ? ? ? 2ֶश͸ใुΛۮવൃݟ͢Δ·Ͱ͸ ϥϯμϜͳࢼߦ͔͠ग़དྷͳ͍ ਪఆ ਪఆ ใु͸ΰʔϧ͚ͩͰͳ͘ ͦ͜·ͰͷಓఔΛ༠ಋ͢ΔΑ͏ʹ ഑ஔ͞Ε͍ͯΔඞཁ͕͋Δ ?

Slide 98

Slide 98 text

͜Ε͸ใु͕ԕ্͗ͯ͢ख͍͔͘ͳ͍ ·ͣԿॲΛ౿ΜͰ໨ඪʹͨͲΓண͔͚ͩ͘Λ൑அ͢Δ ݱࡏ஍ ஍ܗ ໨ඪ ࢟੎ ԿॲΛ౿Ή͔ ʮา͘ҝͷ଍ͷಈ͔͠ํ͕ະ஌ͷঢ়ଶ͔Β໨ඪ஍఺·Ͱา͚ͨΒใुʯ

Slide 99

Slide 99 text

ݱࡏ஍ ஍ܗ ໨ඪ ࢟੎ ԿॲΛ౿Ή͔ ࣍ͷҰาͷҝʹͲ͏ಈ͔͘ ࣍ͷҰาΛ ࢦఆ͞ΕͨҐஔͰ౿ΊͨΒ ใु ໨ඪҐஔ·Ͱ า͘͜ͱ͕Ͱ͖ͨΒ ใु ҰาΛਖ਼͘͠า͘͜ͱΛֶश͢ΔڧԽֶशͱ ໨ඪ஍఺·ͰͲͷΑ͏ͳܦ࿏Ͱา͔͘Λֶश͢ΔڧԽֶशΛ ૊Έ߹ΘͤΔ͜ͱͰ ໨ඪ஍఺·Ͱา͘ҝͷಈ͖ํΛशಘͤ͞Δ ඵʹճ ඵʹճ

Slide 100

Slide 100 text

ڧԽֶशʹΑΔө૾͔Βͷಈ͖ͷशಘ Xue Bin Peng, Angjoo Kanazawa, Jitendra Malik, Pieter Abbeel, and Sergey Levine. 2018. SFV: reinforcement learning of physical skills from videos. In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 178, 14 pages. DOI: https://doi.org/10.1145/3272127.3275014

Slide 101

Slide 101 text

Xue Bin Peng, Angjoo Kanazawa, Jitendra Malik, Pieter Abbeel, and Sergey Levine. 2018. SFV: reinforcement learning of physical skills from videos. In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 178, 14 pages. DOI: https://doi.org/10.1145/3272127.3275014 ө૾͔Β࢟੎Λਪఆ͠ ͓खຊͱͳΔಈ͖Λ࡞Δ ΩϟϥΫλʔ͕෺ཧతʹՄೳͳൣғͰ͓खຊʹ͍ۙಈ͖Λֶश͢Δ ෺ཧγϛϡϨʔγϣϯԼͰ ಈ͘ΩϟϥΫλʔ͕ ֤ϑϨʔϜʹ͓͍ͯ ͓खຊʹ͍ۙϙʔζΛ͍ͯ͠Δఔ ߴ͍ใु

Slide 102

Slide 102 text

ڧԽֶशʹΑΔ෰ΛணΔಈ͖ͷशಘ Alexander Clegg, Wenhao Yu, Jie Tan, C. Karen Liu, and Greg Turk. 2018. Learning to dress: synthesizing human dressing motion via deep reinforcement learning. In SIGGRAPH Asia 2018 Technical Papers (SIGGRAPH Asia '18). ACM, New York, NY, USA, Article 179, 10 pages. DOI: https://doi.org/10.1145/3272127.3275048 ෰ʹ͸෍γϛϡϨʔγϣϯΛߦ͍͍ͨ ࣄલʹΩϟϓνϟͨ͠ಈ͖Ͱ͸ ର৅ͷΩϟϥΫλʔ͸ ਖ਼͘͠෰ΛணΕͳ͍͔΋͠Εͳ͍ https://www.cc.gatech.edu/~aclegg3/projects/LearningToDress.html ڧԽֶशͰΩϟϥΫλʔʹ ࣗྗͰ෰ΛணΔํ๏Λֶशͯ͠΋Β͏

Slide 103

Slide 103 text

r(s) = w1 rp (s) + w2 rd (s) + w3 rg (s) + w4 rt (s) + w5 rr (s) ෰ΛணΔڧԽֶशͷใु rp (s) ਐḿใुକΛ௨աͨ͠࿹΍಄ͷ௕͞ʹൺྫ͢Δใु rd (s) rg (s) rr (s) ͱΓ͋͑ͣକʹ࿹Λ௨ͦ͏ͱ͢ΔΑ͏ʹ༠ಋ มܗใु෰͕ҾͬுΒΕͯ৳ͼ͍ͯΔ΄Ͳݮগ͢Δใु ෰Λഁ͘Α͏ͳಈ͖Λආ͚ΔΑ͏ʹ༠ಋ

Slide 104

Slide 104 text

r(s) = w1 rp (s) + w2 rd (s) + w3 rg (s) + w4 rt (s) + w5 rr (s) ෰ΛணΔڧԽֶशͷใु rp (s) ਐḿใुକΛ௨աͨ͠࿹΍಄ͷ௕͞ʹൺྫ͢Δใु rd (s) มܗใु෰͕ҾͬுΒΕͯ৳ͼ͍ͯΔ΄Ͳݮগ͢Δใु rg (s) ଌ஍ใु෰͕ഽʹ઀৮͍ͯ͠Δׂ߹ʹൺྫ͢Δใु rr (s) ͱΓ͋͑ͣକʹ࿹Λ௨ͦ͏ͱ͢ΔΑ͏ʹ༠ಋ ෰Λഁ͘Α͏ͳಈ͖Λආ͚ΔΑ͏ʹ༠ಋ ෰Λ਎ʹ͚͍ͭͯΔঢ়ଶΛ໨ࢦ͢Α͏ʹ༠ಋ ΰʔϧใु෰Λணͨঢ়ଶʹࣅ͍ͯΔ΄Ͳ૿Ճ͢Δใु କ͔Β಄Λग़ͦ͏ͱͨ͠Γ͠ͳ͍Α͏ʹ༠ಋ ෰ͷछྨʹΑͬͯͷׂ߹Λௐ੔ w

Slide 105

Slide 105 text

ڧԽֶश͕༏ΕͨߦಈΛੜ੒͢ΔͨΊ͸ ใुͷઃܭ͕伴ʹͳΔ ͦΕҎ֎͍͡ΕΔͱ͜Ζ͕ແ͍ͷͰ ใुΛͲͷΑ͏ʹઃఆͯ͠ձಘͯ͠ཉ͍͠ߦಈʹ༠ಋ͢Δ͔ʹͳΔ ͱ΋ݴ͑Δ

Slide 106

Slide 106 text

ϦΞϧλΠϜϨΠτϨʔγϯά ޫͷ෺ཧతͳৼΔ෣͍ΛγϛϡϨʔτ͢ΔࣄͰ ֤ϐΫηϧͷ৭Λܾఆ͢Δ ͋Δ఺͕์ࣹ͢ΔޫΤωϧΪʔ͸ ͦͷ఺͕ࣗΒൃͨ͠ޫΤωϧΪʔͱ ͋ΒΏΔํ޲͔Βͦͷ఺ʹ౸ୡͨ͠ޫΤωϧΪʔͷ૯࿨͔Β ͦͷ఺͕ٵऩͨ͠ޫΤωϧΪʔΛҾ͍ͨ΋ͷʹͳΔ ͜ͷܭࢉʹ͸͋ΒΏΔํ޲ʹ͍ͭͯͷੵ෼͕܁Γฦ͠ݱΕΔ ϞϯςΧϧϩੵ෼Ͱղ͘ͷ͕Ұൠత͕ͩ ϦΞϧλΠϜͱͳΔͱे෼ͳαϯϓϧ਺Λ֬อͰ͖ͳ͍ͨΊ ϊΠζͩΒ͚ͷϨϯμϦϯά݁Ռ͕ग़དྷ্͕Δ

Slide 107

Slide 107 text

ϊΠζͩΒ͚ͷϨϯμϦϯά݁ՌΛΦʔτΤϯίʔμʹ௨ͯ͠ ϊΠζ͕ແ͔ͬͨ৔߹ͷը૾Λਪఆͤ͞Δ Chakravarty R. Alla Chaitanya, Anton S. Kaplanyan, Christoph Schied, Marco Salvi, Aaron Lefohn, Derek Nowrouzezahrai, and Timo Aila. 2017. Interactive reconstruction of Monte Carlo image sequences using a recurrent denoising autoencoder. ACM Trans. Graph. 36, 4, Article 98 (July 2017), 12 pages. DOI: https://doi.org/10.1145/3072959.3073601 https://research.nvidia.com/publication/interactive-reconstruction-monte-carlo-image-sequences-using-recurrent-denoising

Slide 108

Slide 108 text

ࣗಈੜ੒͸༩͑ΒΕͨ৚݅ʹͰ͖Δ͚͍ͩۙ෼෍Λอͱ͏ͱ͢Δ ਓ͕࡞඼Λ໘ന͍ͱײ͡Δͷ͸༧૝Λཪ੾ΒΕͨ࣌Ͱ͋Δ ݱࡏͷػցֶश͸ήʔϜ։ൃΛܶతʹָʹ͢Δ͕ ήʔϜͷ໘നͦ͞ͷ΋ͷΛੜΈग़͢ʹ͸ࢸ͍ͬͯͳ͍

Slide 109

Slide 109 text

ήʔϜΛߏ੒͢Δཁૉ͕ ήʔϜʹݱ࣮ຯΛ༩͑ΔͨΊʹ࡞Δ෺ͳͷ͔ ήʔϜΛ໘ന͘͢ΔͨΊʹ࡞Δ΋ͷͳͷ͔ Λߟ͑Α͏ લऀ͸ػցֶशͰੜ੒ͨ͠ํ͕ྑ͍͔΋͠Εͳ͍ ޙऀ͸ਓ͕ؒ࡞ͬͨํ͕ྑ͍͔΋͠Εͳ͍