모바일 얼굴인식 SDK 개발

#FacialRecognition

딥러닝 기반의 얼굴인식 모델을 처리 성능이 빈약한 모바일 및 전용장비에 적용하기 위한 최적화 방법에 대해서 공유하고자 합니다.

발표자 : andrew.js
카카오엔터프라이즈 비전인텔리전스팀의 앤드루와 포입니다. 얼굴인식을 모바일에 적용하는 일을 하고 있습니다.

po.ai
카카오엔터프라이즈 비전인텔리전스팀의 앤드루와 포입니다. 얼굴인식을 모바일에 적용하는 일을 하고 있습니다.

Transcript

1. $PQZSJHIU
    ,BLBP $PSQ
   "MM
   SJHIUT
   SFTFSWFE
   3FEJTUSJCVUJPO
   PS
   QVCMJD
   EJTQMBZ
   JT
   OPU
   QFSNJUUFE
   XJUIPVU
   XSJUUFO
   QFSNJTTJPO
   GSPN
   ,BLBP ݽ߄ੌ ঴ҷੋध 4%,ѐߊ JG LBLBP  ஠஠য় ূఠ೐ۄ੉ૉ ন੿ࢳ "OESFXKT
   ࣠੿Ӕ 1PBJ
2. ࠗ  ঴ҷ ੋध 4%,
   ࣗѐ +BOVT
   4%,
3. ,BLBP
   &OUFSQSJTF
   ࢎޖप ਕ௼झܖ
4. ݽ߄ੌ ঴ҷੋध 4%,੄ ೙ਃࢿ ҃۝ച ߂ ࢿמ ӓ؀ച ೙ਃ  '375
   ,*04,
   ࠗ࠙ਤী ੸ਊػ ݽ؛਷ ݽ߄ੌী ੸ਊ ೞӝী ցޖ ௼׮  ޷Ҵ Ҵ݀಴ળӝࣿোҳࣗ /*45 ੄ ঴ҷ ੋध ߮؊ ߮஖݃௼ పझ౟
5. ݽ߄ੌ ঴ҷੋध 4%,੄ ೙ਃࢿ ׮নೠ ೒ۖಬ ߂ ٣߄੉झী ੸ਊ оמ  п ٣߄੉झо о૑Ҋ ੓ח ୭Ҋ੄ ࢿמਵ۽ ز੘ ೧ঠ ೠ׮  ৈ۞ ٣߄੉झܳ ഒ೤ ਍৔੉ оמೞ׮ (BMBYZ
   5BC  "OESPJE  4OBQESBHPO ઱ ਬפৡழޭפ౭  "OESPJE  *ӝ߈ /16 "QQMF
   J1BE  J04  Gӝ߈ /16 (BMBYZ
   QIPOF  BOESPJE  &YZOPT
6. ݽ߄ੌ ঴ҷੋध 4%,੄ ೙ਃࢿ ׮নೠ ೒ۖಬ ߂ ٣߄੉झী ੸ਊ оמ  п ٣߄੉झо о૑Ҋ ੓ח ୭Ҋ੄ ࢿמਵ۽ ز੘ ೧ঠ ೠ׮  ৈ۞ ٣߄੉झܳ ഒ೤ ਍৔੉ оמೞ׮ (BMBYZ
   5BC  "OESPJE  4OBQESBHPO ઱ ਬפৡழޭפ౭  "OESPJE  *ӝ߈ /16 "QQMF
   J1BE  J04  Gӝ߈ /16 (BMBYZ
   QIPOF  BOESPJE  &YZOPT
7. ઱ਃ ٣߄੉झ߹ प೯ ࣘب 
   GQT
   ੿بܳ ࢲ࠺झী ਗഝೠ ࣘب۽ ࠅ ࣻ ੓णפ׮ ࢖ࢿ੹੗ (BMBYZ
   5BC 4   GQT ઱ ਬפৡழޭפ౭ 6#JP9
   'BDF
   1SFNJVN  
   GQT "QQMF
   *OD J1BE
    UI HFO   GQT
8. ઱ਃ ٣߄੉झ߹ प೯ ࣘب 
   GQT
   ੿بܳ ࢲ࠺झী ਗഝೠ ࣘب۽ ࠅ ࣻ ੓णפ׮ ࢖ࢿ੹੗ (BMBYZ
   5BC 4   GQT ઱ ਬפৡழޭפ౭ 6#JP9
   'BDF
   1SFNJVN  
   GQT "QQMF
   *OD J1BE
    UI HFO   GQT
9. ࢲߡ "1* ױ݈ "1* 4%,о ૑ਗ ೞח 1MBUGPSN $
   $PSF
   -JC "OESPJE ""3
   -JCSBSZ J04 'SBNFXPSL
   -JCSBSZ 1ZUIPO .PEVMF +BWB
   ,PUMJO +"3
   -JCSBSZ ݽ߄ੌҗ ࢲߡജ҃ী ೙ਃೠ -JCSBSZઁҕ  ݽ߄ੌҗ ࢲߡ জ ࢚ഐ ਍৔੉ ੌ߈੸ ੉׮  ݽ߄ੌ জ ױة ਍৔ ҳࢿ оמ  ࢲߡ জ ױة ਍৔ ҳࢿ оמ ݽ߄ੌ জ .PCJMF
   "QQ ࢲߡ জ 4FSWFS
   "QQ
10. ࢲߡ "1* ױ݈ "1* 4%,о ૑ਗ ೞח 1MBUGPSN $
    $PSF
    -JC "OESPJE ""3
    -JCSBSZ J04 'SBNFXPSL
    -JCSBSZ 1ZUIPO .PEVMF +BWB
    ,PUMJO +"3
    -JCSBSZ ݽ߄ੌҗ ࢲߡജ҃ী ೙ਃೠ -JCSBSZઁҕ  ݽ߄ੌҗ ࢲߡ জ ࢚ഐ ਍৔੉ ੌ߈੸ ੉׮  ݽ߄ੌ জ ױة ਍৔ ҳࢿ оמ  ࢲߡ জ ױة ਍৔ ҳࢿ оמ ݽ߄ੌ জ .PCJMF
    "QQ ࢲߡ জ 4FSWFS
    "QQ
11. ࢲߡ "1* ױ݈ "1* 4%,о ૑ਗ ೞח 1MBUGPSN $
    $PSF
    -JC "OESPJE ""3
    -JCSBSZ J04 'SBNFXPSL
    -JCSBSZ 1ZUIPO .PEVMF +BWB
    ,PUMJO +"3
    -JCSBSZ ݽ߄ੌҗ ࢲߡജ҃ী ೙ਃೠ -JCSBSZઁҕ  ݽ߄ੌҗ ࢲߡ জ ࢚ഐ ਍৔੉ ੌ߈੸ ੉׮  ݽ߄ੌ জ ױة ਍৔ ҳࢿ оמ  ࢲߡ জ ױة ਍৔ ҳࢿ оמ ݽ߄ੌ জ .PCJMF
    "QQ ࢲߡ জ 4FSWFS
    "QQ
12. ࠗ   1MBUGPSN
    -FWFMীࢲ੄ ബਯച ౲
    
13. ঴ҷ ੋध 4%,੄ ݽٕ 1JQFMJOF
14. ঴ҷ ੋध 4%,੄ ݽٕ 1JQFMJOF
15. ٩۞׬ ೐ۨ੐ਕ௼ %- 'SBNFXPSL 1ZUPSDI 0//9 ,FSBT 5FOTPSGMPX-JUF $POWFSUFS 5FOTPSGMPX-JUF *OUFSQSFUFS 5FOTPSGMPX-JUF 'PSNBU ݽ؛ ѐߊ ױ҅ ߓನ ױ҅ $PSFNM
    NPEFM $PSF.- $PSFNM UPPMT
16. ٩۞׬ ೐ۨ੐ਕ௼ %- 'SBNFXPSL 1ZUPSDI 0//9 ,FSBT 5FOTPSGMPX-JUF $POWFSUFS 5FOTPSGMPX-JUF *OUFSQSFUFS 5FOTPSGMPX-JUF 'PSNBU ݽ؛ ѐߊ ױ҅ ߓನ ױ҅ $PSFNM
    NPEFM $PSF.- $PSFNM UPPMT
17. ٩۞׬ ೐ۨ੐ਕ௼ %- 'SBNFXPSL 1ZUPSDI 0//9 ,FSBT 5FOTPSGMPX-JUF $POWFSUFS 5FOTPSGMPX-JUF *OUFSQSFUFS 5FOTPSGMPX-JUF 'PSNBU ݽ؛ ѐߊ ױ҅ ߓನ ױ҅ $PSFNM
    NPEFM $PSF.- $PSFNM UPPMT
18. 5FOTPSGMPX-JUFܳ ݽ߄ੌ ೐ۨ੐ਕ௼۽ ࢶఖೠ ੉ਬ  -BUFODZ ࢿמ੉ જ׮  ׮নೠ ߔূ٘ #BDLFOE ૑ਗ  DQV
    HQV
    OOBQJ
    YOOQBDL
    DPSFNM FUD  "SNӝ߈ Gীࢲח YOOQBDLӝ߈ CBDLFOEо જ਷ ࢿמ  ୭Ӕ੄ %FWJDFীࢲ HQV ӝ߈ ࢿמ੉ ੼੼ જই ૑Ҋ੓׮   //"1* /FVSBM
    /FUXPSL
    "1* ח উ٘۽੉٘੄ ಴ળ ݠन۞׬ झಖ੉׮  ࢚؀੸ਵ۽ ߊৌী ஂডೞ૑ ঋ׮  উ੿੸ੋ ন੗ച 2VBOUJ[BUJPO ૑ਗ  ׮নೠ ೒ۖಬ J04
    "OESPJE
    4FSWFS ૑ਗ ٩۞׬ ೐ۨ੐ਕ௼ %- 'SBNFXPSL
19. 5FOTPSGMPX-JUFܳ ݽ߄ੌ ೐ۨ੐ਕ௼۽ ࢶఖೠ ੉ਬ  -BUFODZ ࢿמ੉ જ׮  ׮নೠ ߔূ٘ #BDLFOE ૑ਗ  DQV
    HQV
    OOBQJ
    YOOQBDL
    DPSFNM FUD  "SNӝ߈ Gীࢲח YOOQBDLӝ߈ CBDLFOEо જ਷ ࢿמ  ୭Ӕ੄ %FWJDFীࢲ HQV ӝ߈ ࢿמ੉ ੼੼ જই ૑Ҋ੓׮   //"1* /FVSBM
    /FUXPSL
    "1* ח উ٘۽੉٘੄ ಴ળ ݠन۞׬ झಖ੉׮  ࢚؀੸ਵ۽ ߊৌী ஂডೞ૑ ঋ׮  উ੿੸ੋ ন੗ച 2VBOUJ[BUJPO ૑ਗ  ׮নೠ ೒ۖಬ J04
    "OESPJE
    4FSWFS ૑ਗ ٩۞׬ ೐ۨ੐ਕ௼ %- 'SBNFXPSL
20. 5FOTPSGMPX-JUFܳ ݽ߄ੌ ೐ۨ੐ਕ௼۽ ࢶఖೠ ੉ਬ  -BUFODZ ࢿמ੉ જ׮  ׮নೠ ߔূ٘ #BDLFOE ૑ਗ  DQV
    HQV
    OOBQJ
    YOOQBDL
    DPSFNM FUD  "SNӝ߈ Gীࢲח YOOQBDLӝ߈ CBDLFOEо જ਷ ࢿמ  ୭Ӕ੄ %FWJDFীࢲ HQV ӝ߈ ࢿמ੉ ੼੼ જই ૑Ҋ੓׮   //"1* /FVSBM
    /FUXPSL
    "1* ח উ٘۽੉٘੄ ಴ળ ݠन۞׬ झಖ੉׮  ࢚؀੸ਵ۽ ߊৌী ஂডೞ૑ ঋ׮  উ੿੸ੋ ন੗ച 2VBOUJ[BUJPO ૑ਗ  ׮নೠ ೒ۖಬ J04
    "OESPJE
    4FSWFS ૑ਗ ٩۞׬ ೐ۨ੐ਕ௼ %- 'SBNFXPSL
21. %- ݽ؛੄ ѐࣻܳ ઴ৈۄ ੋध /FUXPSLҗ ݃झ௼ Ѩ୹ /FUXPSLਸ ా೤  Ѣ੄ ݃झ௼ Ѩ୹ /FUXPSL ݅ఀ੄ -BUFODZਸ ੺ড ೡ ࣻ ੓঻׮  ୶о ௏झ౟о Ѣ੄ হ੉ ׮ܲ ࣘࢿٜب ୶о ೡ ࣻ ੓঻׮
22. %- ݽ؛੄ ѐࣻܳ ઴ৈۄ ੋध /FUXPSLҗ ݃झ௼ Ѩ୹ /FUXPSLਸ ా೤  Ѣ੄ ݃झ௼ Ѩ୹ /FUXPSL ݅ఀ੄ -BUFODZਸ ੺ড ೡ ࣻ ੓঻׮  ୶о ௏झ౟о Ѣ੄ হ੉ ׮ܲ ࣘࢿٜب ୶о ೡ ࣻ ੓঻׮
23. /&0/ਸ ੸ӓ੸ਵ۽ ഝਊೞۄ /&0/਷ "SN੉ ૑ਗೞח 4*.% 4JOHMF
    *OTUSVDUJPO
    .VMUJQMF
    %BUB 2 2 2 4PVSDF
    3FHJTUFS %FTUJOBUJPO 3FHJTUFS 7"%%2G
24. /&0/ਸ ੉޷૑ ୊ܻ *NBHF
    1SPDFTTJOH ী ੸ਊ Image(YUV) Image(RGB) (BMBYZ
    4 6CJP9 /&0/
    ੹ 
    N4 
    N4 /&0/
    റ 
    N4 
    N4 
    $PMPS
    NPEFM
    DPOWFSTJPO 
    3PUBUF
    
    
    'MJQ (BMBYZ
    4 6CJP9 ঴ҷ ৔৉ Ѩ୹ӝ ੹  N4 
    N4 റ 
    N4 
    N4 ঴ҷ ے٘݃௼ Ѩ୹ӝ ੹ 
    N4 
    N4 റ 
    N4 
    N4 ঴ҷ ਤઑ х૑ӝ ੹ 
    N4 
    N4 റ 
    N4 
    N4 ౠ૚ ࢤࢿӝ ੹ 
    N4 
    N4 റ 
    N4 
    N4 
    1JYFM
    /PSNBMJ[BUJPO   
    à  
    PS
       N4ب ࣗ઺೤פ׮
25. /&0/ਸ ੉޷૑ ୊ܻ *NBHF
    1SPDFTTJOH ী ੸ਊ Image(YUV) Image(RGB) (BMBYZ
    4 6CJP9 /&0/
    ੹ 
    N4 
    N4 /&0/
    റ 
    N4 
    N4 
    $PMPS
    NPEFM
    DPOWFSTJPO 
    3PUBUF
    
    
    'MJQ (BMBYZ
    4 6CJP9 ঴ҷ ৔৉ Ѩ୹ӝ ੹  N4 
    N4 റ 
    N4 
    N4 ঴ҷ ے٘݃௼ Ѩ୹ӝ ੹ 
    N4 
    N4 റ 
    N4 
    N4 ঴ҷ ਤઑ х૑ӝ ੹ 
    N4 
    N4 റ 
    N4 
    N4 ౠ૚ ࢤࢿӝ ੹ 
    N4 
    N4 റ 
    N4 
    N4 
    1JYFM
    /PSNBMJ[BUJPO   
    à  
    PS
       N4ب ࣗ઺೤פ׮
26. 1JYFM
    /PSNBMJ[BUJPO
    ୭੸ച ా೤ оמೠ ো࢑ਸ ޷ܻ ࣻ೯ 0GGMJOF  #BUDI /PSNBMJ[BUJPO
    -BZFS੄ GVTJOHҗ ਬࢎ  2VBOUJ[BUJPOਸ Ҋ۰ೞৈ ࣗࣻ੼੉ ߊࢤೞ૑ ঋח NFBO 
    TUE  ਸ ࢎਊೣ ੉ ޷ ૑ QJYFM@NFBO ࡐӝ QJYFM@TUE ա־ӝ QJYFM@NFBO ࡐӝ 8
    
    8
    
    TUE ੉ ޷ ૑ ೟ण җ੿ 5SBJOJOH
    5JNF ѐࢶ ੹ *OGFSFODF
    5JNF ѐࢶ റ *OGFSFODF
    5JNF
27. 1JYFM
    /PSNBMJ[BUJPO
    ୭੸ച ా೤ оמೠ ো࢑ਸ ޷ܻ ࣻ೯ 0GGMJOF  #BUDI /PSNBMJ[BUJPO
    -BZFS੄ GVTJOHҗ ਬࢎ  2VBOUJ[BUJPOਸ Ҋ۰ೞৈ ࣗࣻ੼੉ ߊࢤೞ૑ ঋח NFBO 
    TUE  ਸ ࢎਊೣ ੉ ޷ ૑ QJYFM@NFBO ࡐӝ QJYFM@TUE ա־ӝ QJYFM@NFBO ࡐӝ 8
    
    8
    
    TUE ੉ ޷ ૑ ೟ण җ੿ 5SBJOJOH
    5JNF ѐࢶ ੹ *OGFSFODF
    5JNF ѐࢶ റ *OGFSFODF
    5JNF
28. ࠗ  ݽ؛ ন੗ച 2VBOUJ[BUJPO ੸ਊद੄ ౲
29. ݽ؛ ন੗ച 2VBOUJ[BUJPO ী ੄ೠ -BUFODZ ѐࢶ ֙ ੉റ੄ ٣߄੉झীࢲח JOUҗ G੄ ࢿמ ର੉о ഛোೞ׮  (16
    ࠁ׮ب જ׮ model precision cpu xnnpack gpu nnapi Galaxy S7 f32 32.27 19.50 25.42 32.52 i8 24.89 24.68 17.38 24.63 Galaxy S10 f32 22.36 11.97 12.25 29.41 i8 14.26 15.27 13.19 16.15 Galaxy Note20 f32 12.11 9.31 7.50 14.50 i8 5.55 5.20 5.81 6.17 Galaxy S21 f32 10.36 6.95 5.43 18.51 i8 4.1 4.4 5.63 8.09 .PCJMFOFU 7੄ -BUFODZ
    ஏ੿ NJMMJ
    4FDPOET
30. ݽ؛ ন੗ച 2VBOUJ[BUJPO ী ੄ೠ -BUFODZ ѐࢶ ֙ ੉റ੄ ٣߄੉झীࢲח JOUҗ G੄ ࢿמ ର੉о ഛোೞ׮  (16
    ࠁ׮ب જ׮ model precision cpu xnnpack gpu nnapi Galaxy S7 f32 32.27 19.50 25.42 32.52 i8 24.89 24.68 17.38 24.63 Galaxy S10 f32 22.36 11.97 12.25 29.41 i8 14.26 15.27 13.19 16.15 Galaxy Note20 f32 12.11 9.31 7.50 14.50 i8 5.55 5.20 5.81 6.17 Galaxy S21 f32 10.36 6.95 5.43 18.51 i8 4.1 4.4 5.63 8.09 .PCJMFOFU 7੄ -BUFODZ
    ஏ੿ NJMMJ
    4FDPOET
31. ন੗ച 2VBOUJ[BUJPO ী ஘ച੸ ݽ؛ ࢸ҅ о੉٘  ֎౟ਕ௼੄ %FQUIо ծਸ ࣻ۾ ਬܻ  8JEUIо ևਸ ࣻ۾ ਬܻ -BZFS
     -BZFS
     -BZFS
    
32. ন੗ച 2VBOUJ[BUJPO ী ஘ച੸ ݽ؛ ࢸ҅ о੉٘  ֎౟ਕ௼੄ %FQUIо ծਸ ࣻ۾ ਬܻ  8JEUIо ևਸ ࣻ۾ ਬܻ -BZFS
     -BZFS
     -BZFS
    
33. ন੗ച 2VBOUJ[BUJPO ী ஘ച੸ ݽ؛ ࢸ҅ о੉٘  ਋ഥ҃۽ 4IPSU
    $POOFDUJPO ੉ ੓ਸ ࣻ۾ ਬܻೞ׮ -BZFS
     -BZFS
     -BZFS
        
34. ۨ੉য ࠙ೡ۽ ন੗ച ࢿמ ೞۅ ѐࢶ ୹۱ ౠࢿ੉ ׮ܲ ۨ੉য੄ ޛܻ੸ ࠙ೡ੉ ন੗ചী ਬܻೡ ࣻ ੓׮
    1PTU5SBJOJOH
    4QMJU 'VMMZ$POOFDUFE weights<128x256> bias<128> 'VMMZ$POOFDUFE weights<7x128> Bias<7> 4MJDF begin<0> end<6> 4MJDF begin<6> end<7> -PHJTUJD $PODBU 'VMMZ$POOFDUFE weights<128x256> bias<128> 'VMMZ$POOFDUFE weights<6x128> Bias<6> -PHJTUJD $PODBU 'VMMZ$POOFDUFE weights<1x128> Bias<1> 변경 -128 127 -128 127 -128 127 f32 f32 f32
35. ۨ੉য ࠙ೡ۽ ন੗ച ࢿמ ೞۅ ѐࢶ ୹۱ ౠࢿ੉ ׮ܲ ۨ੉য੄ ޛܻ੸ ࠙ೡ੉ ন੗ചী ਬܻೡ ࣻ ੓׮
    1PTU5SBJOJOH
    4QMJU 'VMMZ$POOFDUFE weights<128x256> bias<128> 'VMMZ$POOFDUFE weights<7x128> Bias<7> 4MJDF begin<0> end<6> 4MJDF begin<6> end<7> -PHJTUJD $PODBU 'VMMZ$POOFDUFE weights<128x256> bias<128> 'VMMZ$POOFDUFE weights<6x128> Bias<6> -PHJTUJD $PODBU 'VMMZ$POOFDUFE weights<1x128> Bias<1> 변경 -128 127 -128 127 -128 127 f32 f32 f32
36. ۨ੉য ࠙ೡ۽ ন੗ച ࢿמ ೞۅ ѐࢶ ୹۱ ౠࢿ੉ ׮ܲ ۨ੉য੄ ޛܻ੸ ࠙ೡ੉ ন੗ചী ਬܻೡ ࣻ ੓׮
    1PTU5SBJOJOH
    4QMJU 'VMMZ$POOFDUFE weights<128x256> bias<128> 'VMMZ$POOFDUFE weights<7x128> Bias<7> 4MJDF begin<0> end<6> 4MJDF begin<6> end<7> -PHJTUJD $PODBU 'VMMZ$POOFDUFE weights<128x256> bias<128> 'VMMZ$POOFDUFE weights<6x128> Bias<6> -PHJTUJD $PODBU 'VMMZ$POOFDUFE weights<1x128> Bias<1> 변경 -128 127 -128 127 -128 127 f32 f32 f32
37. .PEFM
    0QUJNJ[BUJPO ࣘب৬ ಌನݢझ ࢎ੉੄ ୭੸੼ ଺ӝ
38. -PX
    -BUFODZ
    )JHI
    1FSGPSNBODF Computation Performance
39. ࠗ  0WFSBMM
    .PEFM
    0QUJNJ[BUJPO
    1SPDFTT
40. 0WFSBMM
    1SPDFTT 4FMFDU
    5BSHFU
    %FWJDF 4FBSDI
    "SDIJUFDUVSF 1FSGPSNBODF
    0QUJNJ[BUJPO
41. 0WFSBMM
    1SPDFTT 4FMFDU
    5BSHFU
    %FWJDF 4FBSDI
    "SDIJUFDUVSF 1FSGPSNBODF
    0QUJNJ[BUJPO
42. ࠗ  $PNNPO
    .PEFM
    0QUJNJ[BUJPO
    )FVSJTUJDT
43. 4JNQMF
    $PNQVUBUJPOBM
    (SBQI 3FEVDF
    $PNQVUBUJPO 6TF
    0QUJNJ[FE
    0QT
44. 4JNQMF
    $PNQVUBUJPOBM
    (SBQI 3FEVDF
    $PNQVUBUJPO 6TF
    0QUJNJ[FE
    0QT
45. ࠗ  %FWJDFBXBSF
    .PEFM 0QUJNJ[BUJPO
    )FVSJTUJDT
46. $16
    *OGFSFODF 10 15 20 25 30 120 180 240 300 Latency(ms) FLOPS(M) Lower FLOPS, Lower Latency
47. (16/16
    *OGFSFODF Lower Complexity, Lower Latency Better Worse
48. ࠗ  %FTJHO
    4USVDUVSF
49. "SDIJUFDUVSF੄ ҳࢿ ਃࣗ
50. %FTJHO input output depth width resolution
51. .JDSPBSDIJUFDUVSF
52. %FTJHO
    PG
    "SDIJUFDUVSF
53. 4MJN'BDF %FTJHO  'BDF
    3FDPHOJUJPO਷ ҕѐػ QSFUSBJOFE
    CBDLCPOFٜ੄ ೟ण ജ҃җ ௼ѱ ׮ܰ׮  ୭੸੄ EFTJHOਸ ਤ೧ࢲח → Re-design
54. 4MJN'BDF %FTJHO most effective receptive field most effective receptive field Total Capacity input output 1x 2x 4x 8x 1x 2x 4x 8x Receptive Field의 디자인이 전체 디자인의 performance를 결정
55. ࠗ  %FTJHO
    .JDSPBSDIJUFDUVSF
56. .JDSPBSDIJUFDUVSF
    %FTJHO
57. 5XP5SBDLT
58. SimpleFace NAS
59. 'PS
    $16
    /"4 /FVSBM
    "SDIJUFDUVSF
    4FBSDI  /"4ۆ  ઱য૓ ܻࣗझ ؘ੉ఠ కझ௼ী ୭੸ചػ PQFSBUJPOҗ TUSVDUVSFܳ ঌҊ્ܻਵ۽ ఐ࢝  DPNQVUBUJPOBM
    HSBQI੄ п OPEFীࢲ যڃ PQFSBUJPOਸ ࢎਊ೧ ׮਺ OPEF۽ ૓೯ೡ૑ Ѿ੿
60. 'PS
    $16
    /"4 /FVSBM
    "SDIJUFDUVSF
    4FBSDI  1/" 1SPYZMFTT /FVSBM
    "SDIJUFDUVSF
    "EBQUBUJPO   TFBSDI
    TQBDFо ևয૑ݶ /"4੄ ബਯ਷ хࣗ  ӝઓ IVNBOEFTJHOFE
    NPEVMFীࢲ DPOOFDUJPO੄ ਬޖ݅ਸ TFBSDIೞৈ ࢿמ ೱ࢚ 1) D.G. Kim and H.C. Lee, "Proxyless Neural Architecture Adaptation at Once," in IEEE Access, 2022
61. 'PS
    $16
    /"4 /FVSBM
    "SDIJUFDUVSF
    4FBSDI  1/" 1SPYZMFTT /FVSBM
    "SDIJUFDUVSF
    "EBQUBUJPO   TFBSDI
    TQBDFо ևয૑ݶ /"4੄ ബਯ਷ хࣗ  ӝઓ IVNBOEFTJHOFE
    NPEVMFীࢲ DPOOFDUJPO੄ ਬޖ݅ਸ TFBSDIೞৈ ࢿמ ೱ࢚ 1) D.G. Kim and H.C. Lee, "Proxyless Neural Architecture Adaptation at Once," in IEEE Access, 2022
62. 'PS
    $16
    /"4 /FVSBM
    "SDIJUFDUVSF
    4FBSDI  1/" 1SPYZMFTT /FVSBM
    "EBQUBUJPO  CBTFMJOF ؀࠺ TLJQDPOOFDUJPO
    ѐ ୶о  زੌ ো࢑۝ীࢲ ӝઓ 4MJN'BDF ؀࠺ ੋध पಁਯ  хࣗ
63. 'PS
    (16/16
    4JNQMF'BDF Slower Better Performance Faster Moderate Performance
64. 'PS
    (16/16
    4JNQMF'BDF Training Inference Better Performance Faster
65. 'PS
    (16/16
    4JNQMF'BDF  4MJN'BDF ؀࠺ زੌ ۨ੉ఢदীࢲ ੋध पಁਯ  хࣗ  4MJN'BDF ؀࠺ زੌ ࢿמীࢲ JOGFSFODF
    द ࣘب ߓ ૐо
66. ࠗ  ,OPXMFEHF
    %JTUJMMBUJPO
    UP
    *NQSPWF
    1FSGPSNBODF
67. ,OPXMFEHF
    %JTUJMMBUJPO Teacher(Pre-trained) Student(From scratch) Knowledge Distillation teacher에서 student로 지식(knowledge)을 증류(distill)하여 전달
68. ,OPXMFEHF
    %JTUJMMBUJPO feature extractor(=model) embedding classifier logit 얼굴에 대한 정보를 추상적으로 함축 학습 데이터의 identity를 확률 분포로 표현 knowledge to distill  ঴ҷ ੋध੄ ੹୓ ౵੉೐ۄੋীࢲ ୶୹ೡ LOPXMFEHFܳ Ѿ੿
69. ,OPXMFEHF
    %JTUJMMBUJPO hard soft student의 학습을 방해하지 않기 위해 smoothing을 통해 soft한 지식을 전달
70. ,OPXMFEHF
    %JTUJMMBUJPO  UFBDIFS৬ TUVEFOU੄ ର੉о ௼ݶ TUVEFOUо ઁ؀۽ ߓ਋૑ ޅೞӝ ٸޙী TUVEFOUо ୽࠙൤ ழঠ ೞח ޙઁ  ؊ ֫਷ ࢿמਸ о૓ UFBDIFSࠁ׮ زੌೠ ࢿמਸ о૓ UFBDIFSܳ ࢎਊೞח Ѫ੉ ؊ જ਷ ബҗ  EJTUJMMBUJPOਸ ૓೯ೠ TUVEFOUܳ ੤ର TUVEFOU۽ ॄࢲ ৈ۞ ߣ EJTUJMMBUJPOೡ ҃਋ ࢿמ ೱ࢚ small gap similar performance distill multiple times
71. хࢎ೤פ׮