LINE's 3D Recognition Technology and Future Prospects

Transcript

1. LINEの3D認識技術と今後の展望 LINE CVL Yoshihisa IJIRI

2. > ઐ໳ɿίϯϐϡʔλϏδϣϯɾϩϘςΟΫε > 0VUEPPS
   ొࢁɾεΩʔɾୌ८ΓɾࣸਅࡱӨɾόΠΫτϥΠΞϧɾɾɾ > *OEPPS
   ϐΞϊԋ૗ɾྺ࢙ɾᗉ੡ɾίʔώʔᖿઝɾ΢ΟεΩʔɺΫϥ ϑτϏʔϧɾɾɾ > ೥ΦϜϩϯೖࣾ >

   
   إͷݕग़ೝࣝͷσδΧϝɾܞଳి࿩ɺ؂ࢹΧϝϥԠ༻ > 
   ෺ମݕग़ɾŤŞƄŸƃũŖŢŔƃɾ0$3ͷ'"޲͚঎඼Խ > 
   ͠ͳ΍͔ͳ੍ޚΛ࣮ݱ͢Δࣗ཯ιϑτϩϘοτݚڀਪਐ > 
   Ϧαʔνϕϯνϟʔ্ཱͪ͛ 0.30/
   4*/*$
   9 > ೥-*/&ೖࣾ > $PNQVUFS
   7JTJPO
   -BCͷ্ཱͪ͛ -*/&גࣜձࣾ "*Χϯύχʔ "*։ൃࣨ ࣨ௕ɺ$PNQVUFS
   7JTJPO
   -BC Ϛωʔδϟʔ :PTIJIJTB
   *KJSJ
   1I%

3. "*ٕज़ͷա೤ͿΓ 0 1000 2000 3000 4000 5000 6000 7000 8000

   1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 $713
   ࿦จ౤ߘ਺ͷਪҠ 出展：IEEE digital library各年度proceedings これらを元に独⾃に集計し作成

4. ίϯϐϡʔλϏδϣϯٕज़ͷա೤ͿΓ QVCMJDBUJPO I  /BUVSF   5IF
   /FX
   &OHMBOE
   +PVSOBM
   PG
   .FEJDJOF  

   4DJFODF   *&&&$7'
   $POGFSFODF
   PO
   $PNQVUFS
   7JTJPO
   BOE
   1BUUFSO
   3FDPHOJUJPO   5IF
   -BODFU   "EWBODFE
   .BUFSJBMT   $FMM   /BUVSF
   $PNNVOJDBUJPOT   $IFNJDBM
   3FWJFXT   *OUFSOBUJPOBM
   $POGFSFODF
   PO
   -FBSOJOH
   3FQSFTFOUBUJPOT   +"."   /FVSBM
   *OGPSNBUJPO
   1SPDFTTJOH
   4ZTUFNT   1SPDFFEJOHT
   PG
   UIF
   /BUJPOBM
   "DBEFNZ
   PG
   4DJFODFT   +PVSOBM
   PG
   UIF
   "NFSJDBO
   $IFNJDBM
   4PDJFUZ   "OHFXBOEUF
   $IFNJF   $IFNJDBM
   4PDJFUZ
   3FWJFXT   /VDMFJD
   "DJET
   3FTFBSDI   3FOFXBCMF
   BOE
   4VTUBJOBCMF
   &OFSHZ
   3FWJFXT   +PVSOBM
   PG
   $MJOJDBM
   0ODPMPHZ   1IZTJDBM
   3FWJFX
   -FUUFST  QVCMJDBUJPO I  "EWBODFE
   &OFSHZ
   .BUFSJBMT   /BUVSF
   .FEJDJOF   *OUFSOBUJPOBM
   $POGFSFODF
   PO
   .BDIJOF
   -FBSOJOH   &OFSHZ
   
   &OWJSPONFOUBM
   4DJFODF   "$4
   /BOP   4DJFOUJGJD
   3FQPSUT   &VSPQFBO
   $POGFSFODF
   PO
   $PNQVUFS
   7JTJPO   5IF
   -BODFU
   0ODPMPHZ   "EWBODFE
   'VODUJPOBM
   .BUFSJBMT   1-P4 0/&   *&&&$7'
   *OUFSOBUJPOBM
   $POGFSFODF
   PO
   $PNQVUFS
   7JTJPO   /BUVSF
   (FOFUJDT   +PVSOBM
   PG
   $MFBOFS
   1SPEVDUJPO   /BUVSF
   .BUFSJBMT   4DJFODF
   PG
   5IF
   5PUBM
   &OWJSPONFOU   $JSDVMBUJPO   #.+   +PVSOBM
   PG
   UIF
   "NFSJDBO
   $PMMFHF
   PG
   $BSEJPMPHZ   "QQMJFE
   $BUBMZTJT
   #
   &OWJSPONFOUBM   4DJFODF
   "EWBODFT  出展：Google Scholar

5. 5 LINEの藤原師匠 CVPRやICCVにバンバン通していた （それで私もジョインした！）

6. ͳͥ఺܈ʁ

7. ը૾ ✔ ࠲ඪܥ ✔ ॱং ✔ εέʔϧ ఺܈ ʁ ࠲ඪܥ

   ʁ ॱং ʁ εέʔϧ ఺܈ͷ೉͠͞

8. ՝୊Λ৐Γӽ͑ΔͨΊओʹΞϓϩʔνଘࡏ Point-based: Qi et al. [CVPR 2017] Alternative representation: Sinha

   et al. [ECCV 2016] Voxel based: Wu et al. [CVPR 2015] Image-based: Kanezaki et al. [CVPR 2018]  αΠζ  ϝογϡ͕ඞཁ  લॲཧ͕ඞཁ  ࠲ඪ஋ ؔ਺ͱͯ͠දݱ͢Ε͹Α͍ͷͰ͸ʁ ओͳղੳख๏

9. • ಛ௃ͱͯ͠ωοτϫʔΫͷॏΈΛར༻ • લॲཧʴಛघͳωοτϫʔΫΛ࠾༻͢Δ͜ͱʹΑΓ࠲ඪɾεέʔϧෆมʹ https://github.com/kentfuji/NeuralEmbedding /FVSBM
   *NQMJDJU
   &NCFEEJOH
   GPS
   1PJOU
   $MPVE
   "OBMZTJT <'VKJXBSB
   $713
   >

10. ը૾ ✔ ࠲ඪܥ ✔ ॱং ✔ εέʔϧ ఺܈ ʁ ࠲ඪܥ

    ✔ ॱং ✔ εέʔϧ ఺܈ͷ೉͠͞

11. model chair bed ʜ table   ʜ  

     ʜ  ճసෆมͷ࣮ݱ

12. A Closer Look at Rotation-Invariant Deep Point Cloud Analysis [Li

    and Fujiwara+, ICCV2021]

13. • ओ੒෼෼ੳͷ࣠ͷ૊Έ߹ΘͤͰճసΛࣔ͢΋ͷΛಛఆ • 4FMFDUPSϞδϡʔϧΛఏҊ͠࠷దͳ࢟੎ͷநग़Λ࣮ݱ MLP major network pooling softmax !

    24 3N N 3 A Closer Look at Rotation-Invariant Deep Point Cloud Analysis [Li and Fujiwara+, ICCV2021]

14. ը૾ ✔ ࠲ඪܥ ✔ ॱং ✔ εέʔϧ ఺܈ ✔ ࠲ඪܥ

    ✔ ॱং ✔εέʔϧ ఺܈ͷ೉͠͞

15. ϊΠζͷଘࡏ΍ରԠؔ܎͕ະ஌ͷ৔߹ɽɽɽ

16. 𝐴!! 𝐴!" ⋯ 𝐴!# 𝐴"! 𝐴"" ⋯ 𝐴"# ⋮ 𝐴$!

    ⋮ 𝐴$" ⋱ ⋮ ⋯ 𝐴$# 𝐱 = 𝑏! 𝑏" ⋮ 𝑏$ min 𝐱 𝐀𝐱 − 𝐛 " " ఺܈" C͕༩͑ΒΕͨ৔߹ɼม׵YΛٻΊΔ 順序が必要！ ઢܗճؼ (Linear Regression)

17. ఺܈" C͕༩͑ΒΕͨ৔߹ɼஔ׵ߦྻ1ͱม׵YΛٻΊΔ 同じ点数が必要！ min 𝐱, 𝐏 𝐀𝐱 − 𝐏𝐛 "

    " 𝐴!! 𝐴!" ⋯ 𝐴!# 𝐴"! 𝐴"" ⋯ 𝐴"# ⋮ 𝐴$! ⋮ 𝐴$" ⋱ ⋮ ⋯ 𝐴$# 𝐱 = ⋮ 𝑏$ 𝑏" 𝑏! 𝐏() ∈ {0, 1} 5 ( 𝐏() = 1 5 ) 𝐏() = 1 Shuffled Linear Regression [Ashwin+, 2017]

18. ఺܈" C͕༩͑ΒΕͨ৔߹ɼஔ׵ߦྻ1ͱม׵YΛٻΊΔ 外れ値と順序の特定が可能！ min 𝐱, 𝐏 𝐀𝐱 − 𝐏𝐛 "

    " 𝐴!! 𝐴!" ⋯ 𝐴!# 𝐴"! 𝐴"" ⋯ 𝐴"# ⋮ 𝐴$! ⋮ 𝐴$" ⋱ ⋮ ⋯ 𝐴$# 𝐱 = ⋮ 𝑏$ 𝑏" 𝑏! 𝐏() ∈ {0, 1} 5 ( 𝐏() ≤ 1 5 ) 𝐏() ≤ 1 5 (,) 𝐏() = 𝑘 Generalized Shuffled Linear Regression [Li and Fujiwara+, ICCV2021]

19. • ֊਺མͪͷஔ׵ߦྻ΋ѻ͑Δ b4IVGGMFE
    -JOFBS
    3FHSFTTJPO`
    • ఺܈͚ͩͰͳ͘ը૾ͷಛ௃఺ͳͲʹ΋ରԠՄೳ Source Target GSLR (ours) Feature

    matching w/ RANSAC SLR Source FMAP ICP BCICP ZoomOut-100 GSLR (ours) 94.4% 62.0% 83.4% 52.6% 37.1% (FOFSBMJ[FE
    4IVGGMFE
    -JOFBS
    3FHSFTTJPO <-J
    BOE
    'VKJXBSB
    *$$7
    >

20. ̏DʴTime = Motion Motion + Linguistics = Cmd2motion ͜Ε͔Βɾɾɾ

21. 21 -*/&
    $7-ͷ஫ྗٕज़  $79ٕज़ ࣗવݴޠॲཧ ೖྗ Ի੠ σδλϧ ςΩετ ը૾ಈը

    3(#%5 ݴޠ ςΩετ ը૾ ਤද Ի੠ೝࣝ $7
    
    13 ॲཧ ੜ੒ Ի੠߹੒ $( ςΩετग़ ྗ 5F9ͳͲ ϚϧνϝσΟΞೖྗʹରԠ͢ΔϚϧνϞʔμϧॲཧ"*ٕज़ $7
    Y
    ˓˓ٕज़͕ॏཁʹʂ ʢ$7Λத৺ͱͯ͠Έͨͱ͖ͷϚϧνϞʔμϧ"*ٕज़ͷҙຯͰԬຊࢯ͕$79ٕज़ͱ໋໊ʣ

22. %PDVNFOU
    6OEFSTUBOEJOH
    "*0$3 Semantic Information S-Overtime 50% (count) 1 (unitpric e)

    20,000 (price) 20,000 PBI 1,818 Subtotal 18,181 Total 20,000 Cash 100,000 Change 80,000 Tax Included 10% Image Spatial Dependency Parsing for Semi-Structured Document Information Extraction [Hwang+, ACL2020]

23. 23 -BZPVU
    SFDPHOJUJPO ςΩετͷϨΠΞ΢τΛೝࣝ͢Δ͜ͱͰϑΟʔϧυݕࡧΛՄೳͱ͢Δ

24. #FZPOE
    DVSSFOU
    "*0$3ʜ $IBSBDUFS
    UZQF 5FSNJOPMPHZ (SBNNBS 'PSNMBZPVU 5PQJDTTUZMF %PDVNFOU
    UZQF %PNBJO
    LOPXMFEHF 1VSQPTF

    
    UBTL $VTUPNFS TQFDJGJD
    LOPXMFEHF $PNNPO
    LOPXMFEHF 7JTVBM
    QBUUFSOT $POUFYU
    510 MFWFM
    PGG
    POMZ
    XJUI
    WJTVBM
    QBUUFSOT $PNCJOBUJPO
    XJUI
    /-1 CFDPNFT
    DSVDJBM $IBSBDUFS -BOHVBHF 8PSE

25. 25 岡本さんから 次回以降に紹介！

26. 26 -*/&
    $7-ͷ஫ྗٕज़  $79ٕज़ ࣗવݴޠॲཧ ೖྗ Ի੠ σδλϧ ςΩετ ը૾ಈը

    3(#%5 ݴޠ ςΩετ ը૾ ਤද Ի੠ೝࣝ $7
    
    13 ॲཧ ੜ੒ Ի੠߹੒ $( ςΩετग़ ྗ 5F9ͳͲ ϚϧνϝσΟΞೖྗʹରԠ͢ΔϚϧνϞʔμϧॲཧ"*ٕज़ $7
    Y
    ˓˓ٕज़͕ॏཁʹʂ ʢ$7Λத৺ͱͯ͠Έͨͱ͖ͷϚϧνϞʔμϧ"*ٕज़ͷҙຯͰԬຊࢯ͕$79ٕज़ͱ໋໊ʣ

27. STRICTLY CONFIDENTIAL -*/&
    "*
    $PNQBOZͷ໨ࢦ͢ੈք ʮͻͱʹ΍͍͞͠"*ʯ͕ɺ ੜ׆΍ϏδωεʹજΉ൥Θ͠͞Λղফ͠ɺ ʮ͜Ε͔Βͷ͋ͨΓ·͑ʯΛ૑Γ·͢ɻ "*ΧϯύχʔͰ͸ɺ-*/&ͷ΋ͭ"*ٕज़Λফඅऀ޲͚͔Β๏ਓ޲͚·Ͱ෯޿͘ల։͍ͯ͠·͢ɻ อ༗͢Δٕज़ʹ͸ࣗવݴޠॲཧɺจࣈɺը૾ɺإɺԻ੠ͷೝࣝ΍Ի੠߹੒ͳͲ͕͋Γɺ ࣾձ΍اۀͷ՝୊΍χʔζʹ߹Θͤͯઃܭ͔Β࣮૷·ͰΛߦ͍ɺ"*ͷࣾձਁಁΛਪਐ͍ͯ͠·͢ɻ ͦΜͳࢲͨͪ͸ɺ

    ʮΑΓࣗવͳϢʔβʔମݧΛ -JGF
    PO
    -*/& ʹ΋ͨΒ͢͜ͱͰ ͜Ε͔Βͷ͋ͨΓ·͑Λͭ͘Γͩ͢ʯ ͱ͍͏7JTJPOΛ࣋ͬͯ೔ʑΛա͍ͯ͝͠·͢ɻ Ϗδωεͱ"*ɺਓͱ"*ͷڑ཭Λ͚ۙͮɺ ೔ʑͷۀ຿΍ͦͷઌͷਓʑͷੜ׆ʹدΓఴ͏ʮ͜Ε͔Βͷ͋ͨΓ·͑ʯΛ૑ग़͠ɺ ΑΓศརͳࣾձΛ࣮ݱ͠·͢ɻ

28. STRICTLY CONFIDENTIAL $-07"
    $IBUCPU -*/&͔ΜͨΜϔϧϓ΍$-07"Ͱഓͬͨ ࣗવݴޠٕज़Λɺ'"2΍઀٬༻#PUʹ ల։Ͱ͖ΔαʔϏε LINE CLOVA Chatzbot $-07"
    0$3

    ࠃࡍձٞͰੈք࠷ߴਫ४ͱೝΊΒΕͨ OCRٕज़Λਃࠐॻ΍ྖऩॻͳͲͷಡΈऔΓɺ ࣗಈೖྗʹ׆༻Ͱ͖ΔαʔϏε LINE CLOVA OCR $-07"
    4QFFDI $-07"ͷԻ੠ೝٕࣝज़Λ׆༻͠ɺ ి࿩΍ಈըϝσΟΞͷԻ੠ॻ͖ى͜͠ɺ ి࿩ԠରͷࣗಈԽαʔϏεͳͲΛఏڙ LINE CLOVA Speech $-07"
    7PJDF $-07"ͷԻ੠߹੒ٕज़Λ׆༻͠ɺ اۀϒϥϯυ΍༻్ʹ͋ͬͨԻ੠ϞσϧΛ࡞੒ ͢ΔαʔϏεΛఏڙ༧ఆ LINE CLOVA Voice $-07"
    5FYU
    "OBMZUJDT ςΩετղੳɺײ৘෼ੳٕज़ɻ Ի੠ೝࣝͰىͨ͜͠ςΩετ͔Βͷݕࡧ΍ ײ৘෼ੳͳͲʹ׆༻ɻ LINE CLOVA Text Analytics $-07"
    7JTJPO ෺ମೝࣝɺը૾ೝٕࣝज़ɻ LINEγϣοϐϯάͷʮSHOPPING LENSʯͰ׆༻ɻ LINE CLOVA Vision $-07"
    'BDF ߴਫ਼౓ͷإೝٕࣝज़ɻ eKYCʢΦϯϥΠϯຊਓ֬ೝʣ ΍إೝূʹΑΔड෇ͳͲʹ׆༻ɻ LINE CLOVA Face -*/&
    $-07"
    ͷ ϓϩμΫτ 4BB4ఏڙ 4BB4ఏڙ -*/&ͷ࣋ͭଟ༷ͳ"*ཁૉٕज़Λجʹ෯޿͍##޲͚ϓϩμΫτΛల։͍ͯ͠·͢ʢҰ෦4BB4ͱͯ͠ఏڙʣɻ

29. "*Χϯύχʔ͕ఏڙ͍ͯ͠ΔαʔϏε

30. "*Χϯύχʔ ͷ 3%
    7JTJPO $POTFSWBUJWF %JTSVQUJWF 5JNF    

    *OUFSBDUJWF
    WJSUVBM
    FYQFSJFODF "VUPOPNPVT
    "*
    XPSLGMPX %JHJUBM
    .F .F
    "7"5"3 %JHJUBM
    *EFOUJUZ #FUUFS
    $BSF 5SVTUXPSUIZ
    "* "*
    'BJSOFTT &YQMBJOBCMF
    "* %BSL
    %BUB 0NOJQPUFOU
    "* (JHBOUJD
    -BOHVBHF
    NPEFM 6OMBCFMFE
    %BUB %BUB
    .BSLFUQMBDF (FOFSBUJWF
    *OUFMMJHFODF /FX
    &EVDBUJPO %FQFOEBCMF
    455 1SJWBDZ
    QSFTFSWJOH 4FBN
    %JTDSJNJOBUPS
31. None

32. 
    ʣ
    *$"441
    ʣ*/5&341&&$)
    ʣ8"41""
    ʣ #JH%BUB  ʢʣ
    *$"441
     &64*1$0
    

    
    */5&341&&$)
    
    %$"4&
    
    "14*1"
    
    $713
    
    51%1 
    '-*$.- 
    
    -%3$  
    
    *$"441
    
    *$3"
    
    *6* 
    *$%&
     *$$7  各分野最⾼峰の会議で認められるAI 基礎研究成果 ͜Ε·Ͱͷ੒Ռ

33. ⾃由度が⾼い発話のリアルタイム認識で⾃然な会話の 書き起こしを実現！ ։ൃதͷٕज़ 4QFFDI CLOVA note

34. %// Ի੠߹੒ ʙײ৘Λॊೈʹ੍ޚՄೳͳԻ੠߹੒Λ࣮ݱʙ COntrollable, High-quality, And expRessIve TTS 明るさ 暗さ

    😀 😄 🙂 😐 😢 😰 😥 ։ൃதͷٕज़ʢ7PJDFʣ

35. HyperCLOVA 1750億超のパラメータを有する汎⽤⾔語モデルを開発 ։ൃதͷٕज़ʢ/-1ʣ

36. 国会図書館デジタルアーカイブ プロジェクト 247万点2.23億ページ超のデジタル・アーカイブ化 ։ൃதͷٕज़ʢ$7-ʣ https://linecorp.com/ja/pr/news/ja/2021/3825

37. ੜ׆΍ϏδωεʹજΉ൥Θ͠͞Λղܾ͠ ͜Ε͔Βͷ͋ͨΓ·͑Λ૑Γग़͢ʂ Ұྲྀʹͩ͜ΘΔΠϯλʔφγϣφϧͳνʔϜ

38. None

39. Our challenge Innovation by mixing LINE AI assets, especially NLP,

    voice/speech, and CV .JYFE
    -*/&
    "*
    .J-"* .VMUJNPEBM
    JOQVUPVUQVU
40. None
41. None
42. None
43. None

44. 44 STRICTLY CONFIDENTIAL $-07"
    0$3 Point 1 ੈք࠷ߴਫ४ͷ"*0$3 Point 2 ͋ΒΏΔॻྨ΍ը૾Λૉૣ͘ςΩετԽ

    Point 3 खॻ͖ͷจࣈ΋ೝࣝՄೳ ԣॻ͖΍ॎॻ͖ɺؙ͘࿷ۂͨ͠จࣈͳͲѱ৚݅ԼͰͷಡΈऔΓɺଟݴޠͷ ೝࣝɺઐ໳༻ޠͷೝࣝͳͲͰߴ͍ਫ਼౓ͱධՁɻจॻղੳͱೝࣝʹؔ͢Δࠃ ࡍձٞ *$%"3
     ͷ෼໺ʹͯੈք/PΛ֫ಘ͍ͯ͠·͢ɻ ϑΥʔϚοτ͕ܾ·͍ͬͯΔॻྨ͸΋ͪΖΜɺ͋ΒΏΔελΠϧͷॻྨΛ ਖ਼͘͠ςΩετԽ͠·͢ɻ$-07"
    0$3ʢྖऩॻɾ੥ٻॻɾϨγʔτಛ ԽܕʣͰ͸ɺϑΥʔϚοτͷࣄલొ࿥͕ෆཁɻ खॻ͖จࣈ΍ɺࣼΊʹͳͬͨจࣈ΋ߴਫ਼౓ͷೝূ͕Մೳ

45. 45 STRICTLY CONFIDENTIAL 4"1
    $PVODVS +BQBO ࢴͷ੥ٻॻͷσδλϧԽͷύʔτφʔͱͯ͠-*/&
    $-07"Λબఆ גࣜձࣾതใಊ%:ϝσΟΞύʔτφʔζ γϦΞϧφϯόʔΛಡΈऔΔ͜ͱͰɺίϯϏχԁ͘͡ΛΦϯϥΠϯԽ -*/&τʔΫϧʔϜ τʔΫϧʔϜ͔Βը૾ΛࡱΔ͚ͩͰจࣈೝࣝػೳ͕ར༻Մೳ

    Ϋϥ΢υαʔϏεͱͷύʔτφʔγοϓ ৽ͨͳιϦϡʔγϣϯͱͯ͠ͷ׆༻ -*/&αʔϏε΁ͷߩݙ ʘ GPS
    */70*$&
    ʗ -*/&
    Ϩγʔτ 
    -*/&
    1-"$& ϨγʔτΛ"*ͰಡΈऔΔ͜ͱͰɺֹۚͱ඼໨͕ࣗಈͰ෼ྨɻ ࢧग़؅ཧ΍ར༻͓ͨ͠ళͷޱίϛαΠτ΁ͷ౤ߘͳͲ͕؆୯ʹɻ $-07"
    0$3ಋೖࣄྫ

46. 46 STRICTLY CONFIDENTIAL $POGJEFOUJBM -*/&
    "J$BMM Point 1 ϢʔβʔΛ଴ͨͤͳ͍ར༻ମݧ Point 2

    ਓؒຯ͋;ΕΔࣗવͳର࿩ Point 3 طଘγεςϜ΍-*/&ͱͷ࿈ܞ ࣌ؒ೔ɺडిମ੍Λ༻ҙͰ͖Δ͜ͱ͸΋ͪΖΜɺ൪߸ೖྗͰରԠ ಺༰ΛৼΓ෼͚Δ*73ʢ*OUFSBDUJWF
    7PJDF
    3FTQPOTFʣͱ͸ҟͳΓɺॊೈ ʹରԠ͠·͢ɻ ཈༲ͷ͋Δਓؒʹ͍ۙࣗવͳԻ੠ͰɺϢʔβʔʹετϨεΛֻ͚·ͤΜɻ ·ͨɺ"*ʹΑΔԻ੠ೝࣝͷֶशʹΑΓɺԻ੠ೝࣝͱର࿩ͷਫ਼౓͕޲্͠ɺ ରԠ඼࣭͕޲্͠·͢ɻ ͜Ε·ͰՍిडిޙʹߦ͍ͬͯͨΞφϩάͳσʔλ੔උۀ຿΋ɺγες Ϝ࿈ܞʹΑΓܰݮ͠·͢ɻ·ͨɺ-*/&΍4.4ͱ࿈ܞ͢Δ͜ͱͰɺ௨࿩ޙ ʹϢʔβʔʹࣗಈͰϝοηʔδΛૹ৴͢Δ͜ͱ΋ՄೳͰ͢ɻ

47. 47 STRICTLY CONFIDENTIAL ϠϚτӡ༌גࣜձࣾ ސ٬͔Βͷి࿩ʹΑΔूՙड෇ ਆಸ઒ݝ ৽ܕίϩφి࿩૬ஊ૭ޱ גࣜձࣾΤϏιϧ ҿ৯ళ޲͚༧໿؅ཧγεςϜ גࣜձࣾΧʔϑϩϯςΟΞ

    Χʔϝϯςφϯε༧໿αʔϏε େखاۀͷۀ຿ʹಋೖ ެڞߦ੓ͷෛ୲ܰݮ ϓϥοτϑΥʔϜͱͷػೳ࿈ܞ -*/&
    "J$BMM ಋೖ࣮੷

48. 48 STRICTLY CONFIDENTIAL -*/&
    F,:$ Point ߴਫ਼౓ͷΦϯϥΠϯຊਓ֬ೝ -*/& F,:$͸ɺ-*/&͕։ൃͨ͠"*ٕज़Λ૊Έ߹Θͤɺ҆શੑͱརศੑΛ ཱ྆ͨ͠ɺΦϯϥΠϯ্Ͱͷຊਓ֬ೝΛ׬݁͢ΔιϦϡʔγϣϯͰ͢ɻ "1*΍4%,ͳͲ๛෋ͳఏڙํ๏ʹΑΓɺར༻໨తʹ͋ͬͨΧελϚΠζ͕

    ՄೳͰ͢ɻखଓ͖ͷ؆ུԽʹΑͬͯɺۀ຿ޮ཰ԽɾϢʔβʔ΁ͷརศੑ ఏڙɺͳΓ͢·͠ʹΑΔෆਖ਼ΞΫηεɾෆਖ਼ར༻ͷ๷ࢭΛ࣮ݱ͠·͢ɻ

49. 49 STRICTLY CONFIDENTIAL -*/&
    1BZ εϚϗͱ਎෼ূ͕͋Ε͹Ͱ͖ΔʮεϚϗͰ͔ΜͨΜຊਓ֬ೝʯ -*/&
    1BZͰͷಋೖ -*/&
    F,:$ ಋೖࣄྫ