Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
20170204-meganeco
Search
ARIYAMA Keiji
February 04, 2017
Technology
2
1.6k
20170204-meganeco
「TensorFlowで趣味の画像収集サーバーを作る 2017年2月号」
顔領域検出ネットワークを作成し、さくらの高火力サーバー上で動作させました。
ARIYAMA Keiji
February 04, 2017
Tweet
Share
More Decks by ARIYAMA Keiji
See All by ARIYAMA Keiji
Build with AI
keiji
0
130
DroidKaigi 2023
keiji
0
1.5k
TechFeed Conference 2022
keiji
0
220
Android Bazaar and Conference Diverse 2021 Winter
keiji
0
830
ci-cd-conference-2021
keiji
1
1.2k
Android Bazaar and Conference 2021 Spring
keiji
3
750
TFUG KANSAI 20190928
keiji
0
90
Softpia Japan Seminar 20190724
keiji
1
140
pixiv App Night 20190611
keiji
1
560
Other Decks in Technology
See All in Technology
来年もre:Invent2024 に行きたいあなたへ - “集中”と“つながり”で楽しむ -
ny7760
0
450
「最高のチューニング」をしないために / hack@delta 24.10
fujiwara3
21
3.4k
オニオンアーキテクチャで実現した 本質課題を解決する インフラ移行の実例
hryushm
14
3k
大規模データ基盤チームのオンプレTiDB運用への挑戦 / dpu-tidb
cyberagentdevelopers
PRO
1
110
カメラを用いた店内計測におけるオプトインの仕組みの実現 / ai-optin-camera
cyberagentdevelopers
PRO
1
120
Nix入門パラダイム編
asa1984
2
200
顧客が本当に必要だったもの - パフォーマンス改善編 / Make what is needed
soudai
24
6.7k
生成AIの強みと弱みを理解して、生成AIがもたらすパワーをプロダクトの価値へ繋げるために実践したこと / advance-ai-generating
cyberagentdevelopers
PRO
1
180
Automated Promptingを目指すその前に / Before we can aim for Automated Prompting
rkaga
0
110
Commitment vs Harrisonism - Keynote for Scrum Niseko 2024
miholovesq
6
1k
AWS re:Inventを徹底的に楽しむためのTips / Tips for thoroughly enjoying AWS re:Invent
yuj1osm
1
540
サイロ化した金融システムを、packwerk を利用して無事故でリファクタリングした話
coincheck_recruit
3
3.7k
Featured
See All Featured
What's new in Ruby 2.0
geeforr
342
31k
The Illustrated Children's Guide to Kubernetes
chrisshort
48
48k
Navigating Team Friction
lara
183
14k
Code Reviewing Like a Champion
maltzj
519
39k
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
4
290
Speed Design
sergeychernyshev
24
570
10 Git Anti Patterns You Should be Aware of
lemiorhan
654
59k
Learning to Love Humans: Emotional Interface Design
aarron
272
40k
Measuring & Analyzing Core Web Vitals
bluesmoon
1
39
Design and Strategy: How to Deal with People Who Don’t "Get" Design
morganepeng
126
18k
Statistics for Hackers
jakevdp
796
220k
Making Projects Easy
brettharned
115
5.9k
Transcript
C-LIS CO., LTD.
༗ࢁܓೋʢ,FJKJ"3*:"."ʣ C-LIS CO., LTD. Androidアプリ開発者 機械学習初⼼心者 やってません Photo :
Koji MORIGUCHI (AUN CREATIVE FIRM) と、⾔言いながら早1年年……
ຊΛग़ͤ ιϑτɾڥ͕Ξοϓσʔτ͢Δ
None
None
None
C-LIS CO., LTD. IUUQBN[OUPC,3N 7/29発売
C-LIS CO., LTD.
None
5FOTPS'MPXͰ झຯͷը૾ऩूαʔόʔΛ࡞Δ ݄̎߸
Έͷ؟ڸ່ͬը૾Λ ࣗಈͰऩू͍ͨ͠
4PGUXBSF%FTJHO݄߸ʰ"OESPJE4UVEJPͷελΠϧͰޮΞοϓʱΑΓ ؟ ڸ ͬ ່ ࠜઇΕ͍
ݱࡏͷߏ Downloader Face Detection Megane Detection API 5FOTPS'MPX 確認・修正
認識結果 JSON ܇࿅ λΠϜϥΠϯ ϝσΟΞ データセット ܇࿅
C-LIS CO., LTD. ̎Ϋϥεྨ 1 0
؟ڸ່ͬσʔληοτ
݄ ؟ڸ່ͬը૾ຕ ඇ؟ڸ່ͬຕ ܭɿ ຕ ؟ڸ່ͬ ඇ؟ڸ່ͬ ޡݕग़ ؟ڸ່ͬ
ඇ؟ڸ່ͬ
݄ ؟ڸ່ͬը૾ ຕ ඇ؟ڸ່ͬ ຕ ܭɿ ຕ ؟ڸ່ͬ ඇ؟ڸ່ͬ
ෛྫʢإྖҬͰͳ͍ʣ ຕ
ϞσϧʢωοτϫʔΫʣ
σʔληοτ࡞ʢ5'3FDPSEʣ is_face: 1 is_megane: 1 is_face: 1 is_megane: 0
is_face: 0 is_megane: 0 ࠜઇΕ͍
EFG@DPOWFSU@UGSFDPSET QBUI JT@GBDF JT@NFHBOF PVUQVU@EJS IBTI JOEFY PSJHJOBM@GJMF@OBNF
GJMF@OBNFPTQBUICBTFOBNF QBUI XJUIPQFO QBUI SC BTJNBHF PVUQVU@QBUIPTQBUIKPJO PVUQVU@EJS TEUGSFDPSE IBTI JOEFY XJUIUGQZUIPO@JP5'3FDPSE8SJUFS PVUQVU@QBUI BTXSJUFS FYBNQMFUGUSBJO&YBNQMF GFBUVSFTUGUSBJO'FBUVSFT GFBUVSF\ GJMF@OBNF@CZUFT@GFBUVSF GJMF@OBNFFODPEF PSJHJOBM@GJMF@OBNF@CZUFT@GFBUVSF PSJHJOBM@GJMF@OBNFFODPEF JT@GBDF@JOU@GFBUVSF JT@GBDF JT@NFHBOF@JOU@GFBUVSF JT@NFHBOF JNBHF@CZUFT@GFBUVSF JNBHFSFBE ^ XSJUFSXSJUF FYBNQMF4FSJBMJ[F5P4USJOH σʔληοτ࡞ʢ5'3FDPSEʣ
إೝࣝ༻ͷϞσϧ DPO YY DPO YY GD QPPM Y
DPO YY DPO YY QPPM Y GD PVUQVU MSO MSO Y
SFTIBQFUGSFTIBQF QPPM <CBUDI@TJ[F > EJNSFTIBQFHFU@TIBQF <>WBMVF GD@GD GD
JOQVU@MBZFSSFTIBQF XFJHIUT@TIBQF<EJN > XFJHIU@TUEEFW CJBTFT@TIBQF<> CJBTFT@WBMVF LFFQ@QSPCLFFQ@QSPC શ݁߹ͷೖྗ
܇࿅ 12万ステップ学習後のテストデータによる正答率 96%
؟ڸ່ͬೝࣝ༻ͷϞσϧ 7((/FUΛࢀߟʹΈࠐΈͷ࿈ଓΛ༻͍ͨ $//ʢ$POWPMVUJPOBM/FVSBM/FUXPSLʣϞσϧ DPO YY DPO YY GD
QPPM Y DPO YY DPO YY QPPM Y GD GD PVUQVU
܇࿅ ֶशΞϧΰϦζϜ"EBN ֶश ϛχόον
ݕূ σʔληοτ͔ΒΛςετ༻ͱͯ͠ ਖ਼ʙ
%BUB"VHNFOUBUJPO ը૾Λ ʜ·ͰɺͦΕͧΕճసͨ͠ը૾Λ࡞ ಡΈࠐΈ࣌ͷॲཧ 3BOEPN$SPQʢQYதɺQYྖҬͰ͘Γൈ͘ʣ 3BOEPN'MJQʢԣ࣠ํʹసʣ 3BOEPN6Q%PXOʢॎ࣠ํʹసʣ 3BOEPN#SJHIUOFTT 3BOEPN$POUSBTU
EFG@EJTUPSU JNBHF DSPQ@TJ[F/POF JNBHF@DIBOOFMT CSJHIUOFTT@NBY@EFMUB DPOUSBTU@MPXFS
DPOUSBTU@VQQFS XJUIUGOBNF@TDPQF EJTUPSU BTTDPQF EJTUPSUFE@JNBHFUGDBTU JNBHF UGGMPBU JGDSPQ@TJ[F EJTUPSUFE@JNBHFUGSBOEPN@DSPQ EJTUPSUFE@JNBHF <DSPQ@TJ[F DSPQ@TJ[F > EJTUPSUFE@JNBHFUGJNBHFSBOEPN@GMJQ@MFGU@SJHIU EJTUPSUFE@JNBHF EJTUPSUFE@JNBHFUGJNBHFSBOEPN@GMJQ@VQ@EPXO EJTUPSUFE@JNBHF EJTUPSUFE@JNBHFUGJNBHFSBOEPN@CSJHIUOFTT EJTUPSUFE@JNBHF NBY@EFMUBCSJHIUOFTT@NBY@EFMUB EJTUPSUFE@JNBHFUGJNBHFSBOEPN@DPOUSBTU EJTUPSUFE@JNBHF MPXFSDPOUSBTU@MPXFS VQQFSDPOUSBTU@VQQFS SFUVSOEJTUPSUFE@JNBHF
ʮީิྖҬʯͷΓग़͕͠ඞཁ
C-LIS CO., LTD. ࠜઇΕ͍
C-LIS CO., LTD. ࠜઇΕ͍ 4MJEJOH8JOEPX
C-LIS CO., LTD. 4FMFDUJWF4FBSDI ࠜઇΕ͍
C-LIS CO., LTD. 4FMFDUJWF4FBSDI ʢҾ༻4FMFDUJWF4FBSDIGPS0CKFDU3FDPHOJUJPOIUUQTJWJGOXJVWBOMJTJTQVCMJDBUJPOTCJCUFYCSPXTFSQIQ LFZ6JKMJOHT*+$7CJCBMMCJCʣ
C-LIS CO., LTD. 4FMFDUJWF4FBSDI "MQBDBࣾʹΑΔ࣮ IUUQCMPHKQBMQBDBBJFOUSZ EMJCͷ࣮ IUUQTHJUIVCDPNEBWJTLJOHEMJCCMPCNBTUFS QZUIPO@FYBNQMFTpOE@DBOEJEBUF@PCKFDU@MPDBUJPOTQZ
ݱࡏͷߏ Downloader Face Detection Megane Detection API 5FOTPS'MPX 確認・修正
認識結果 JSON ܇࿅ λΠϜϥΠϯ ϝσΟΞ データセット ܇࿅
إఆʢ'BDF%FUFDUJPOʣͷ Λ্͍͛ͨ
'BTU3$// 'BTUFS3$//
σʔληοτʢ5'3FDPSEʣ ࠜઇΕ͍
ΦʔτΤϯίʔμʢࣗݾූ߸ثʣ DPOW YY DPOW YY DPOW YY DPOW YY
DPOW YY EDPOW YY EDPOW YY EDPOW YY EFDPOW@MBTU YY EDPOW YY ࠜઇΕ͍
EFG@@WBSJBCMF@PO@DQV OBNF TIBQF JOJUJBMJ[FS XJUIUGEFWJDF DQV WBSUGHFU@WBSJBCMF OBNF
TIBQF JOJUJBMJ[FSJOJUJBMJ[FS SFUVSOWBS EFG@HFU@XFJHIUT TIBQF TUEEFW WBS@@WBSJBCMF@PO@DQV XFJHIUT TIBQF UGUSVODBUFE@OPSNBM@JOJUJBMJ[FS TUEEFWTUEEFW SFUVSOWBS EFG@HFU@CJBTFT TIBQF WBMVF WBS@@WBSJBCMF@PO@DQV CJBTFT TIBQF UGDPOTUBOU@JOJUJBMJ[FS WBMVF SFUVSOWBS
EFG@DPOW OBNF JOQVU@MBZFS XFJHIUT@TIBQF XFJHIU@TUEEFW CJBTFT@TIBQF CJBTFT@WBMVF QBEEJOH7"-*%
XJUIUGWBSJBCMF@TDPQF OBNF BTTDPQF XFJHIUT@HFU@XFJHIUT XFJHIUT@TIBQF TUEEFWXFJHIU@TUEEFW CJBTFT@HFU@CJBTFT CJBTFT@TIBQF WBMVFCJBTFT@WBMVF DPOWUGOODPOWE JOQVU@MBZFS XFJHIUT < > QBEEJOHQBEEJOH CJBTUGOOCJBT@BEE DPOW CJBTFT DPOWUGOOSFMV CJBT OBNFTDPQFOBNF @BDUJWBUJPO@TVNNBSZ DPOW SFUVSODPOW
EFG@EDPOW OBNF JOQVU@MBZFS PVUQVU@TIBQFT XFJHIUT@TIBQF XFJHIU@TUEEFW CJBTFT@TIBQF CJBTFT@WBMVF
QBEEJOH7"-*% BDUJWBUF@GVODUJPOSFMV XJUIUGWBSJBCMF@TDPQF OBNF BTTDPQF XFJHIUT@HFU@XFJHIUT XFJHIUT@TIBQF TUEEFWXFJHIU@TUEEFW CJBTFT@HFU@CJBTFT CJBTFT@TIBQF WBMVFCJBTFT@WBMVF EDPOWUGOODPOWE@USBOTQPTF JOQVU@MBZFS XFJHIUT PVUQVU@TIBQFT < > QBEEJOHQBEEJOH CJBTUGOOCJBT@BEE DPOW CJBTFT JGBDUJWBUF@GVODUJPOTJHNPJE EDPOWUGOOTJHNPJE CJBT OBNFTDPQFOBNF FMJGBDUJWBUF@GVODUJPOTPGUTJHO EDPOWUGOOTPGUTJHO CJBT OBNFTDPQFOBNF FMTF EDPOWUGOOSFMV CJBT OBNFTDPQFOBNF @BDUJWBUJPO@TVNNBSZ EDPOW SFUVSOEDPOW
EFGJOGFSFODF JNBHF@OPEF JNBHF@TJ[F DIBOOFMT CBUDI@TJ[F DPOW@DPOW DPOW JOQVU@MBZFSJNBHF@OPEF
XFJHIUT@TIBQF< DIBOOFMT > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% DPOW@DPOW DPOW JOQVU@MBZFSDPOW XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% DPOW@DPOW DPOW JOQVU@MBZFSDPOW XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% DPOW@DPOW DPOW JOQVU@MBZFSDPOW XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% DPOW@DPOW DPOW JOQVU@MBZFSDPOW XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*%
TIBQFTDPOWHFU@TIBQF EDPOW@EDPOW EFDPOW JOQVU@MBZFSDPOW PVUQVU@TIBQFT<CBUDI@TJ[F JOU TIBQFT<>
JOU TIBQFT<> > XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% TIBQFTDPOWHFU@TIBQF EDPOW@EDPOW EFDPOW JOQVU@MBZFSEDPOW PVUQVU@TIBQFT<CBUDI@TJ[F JOU TIBQFT<> JOU TIBQFT<> > XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% TIBQFTDPOWHFU@TIBQF EDPOW@EDPOW EFDPOW JOQVU@MBZFSEDPOW PVUQVU@TIBQFT<CBUDI@TJ[F JOU TIBQFT<> JOU TIBQFT<> > XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*%
TIBQFTDPOWHFU@TIBQF EDPOW@EDPOW EFDPOW JOQVU@MBZFSEDPOW PVUQVU@TIBQFT<CBUDI@TJ[F JOU TIBQFT<>
JOU TIBQFT<> > XFJHIUT@TIBQF< > XFJHIU@TUEEFWF CJBTFT@TIBQF<> CJBTFT@WBMVF QBEEJOH7"-*% EDPOW@MBTU@EDPOW EDPOW@MBTU JOQVU@MBZFSEDPOW PVUQVU@TIBQFT<CBUDI@TJ[F JNBHF@TJ[F JNBHF@TJ[F DIBOOFMT> XFJHIUT@TIBQF< DIBOOFMT > XFJHIU@TUEEFWF CJBTFT@TIBQF<DIBOOFMT> CJBTFT@WBMVF QBEEJOH7"-*% SFUVSOEDPOW@MBTU
σʔληοτʢ5'3FDPSEʣ ࠜઇΕ͍
EFG@@DPOWFSU@UGSFDPSET JNBHF@QBUI NBTL@QBUI PVUQVU@QBUI GJMF@OBNFPTQBUICBTFOBNF JNBHF@QBUI XJUIPQFO
JNBHF@QBUI SC BTJNBHF JNBHF@EBUBJNBHFSFBE XJUIPQFO NBTL@QBUI SC BTJNBHF NBTL@EBUBJNBHFSFBE XJUIUGQZUIPO@JP5'3FDPSE8SJUFS PVUQVU@QBUI BTXSJUFS FYBNQMFUGUSBJO&YBNQMF GFBUVSFTUGUSBJO'FBUVSFT GFBUVSF\ GJMF@OBNF@CZUFT@GFBUVSF GJMF@OBNFFODPEF JNBHF@CZUFT@GFBUVSF JNBHF@EBUB NBTL@CZUFT@GFBUVSF NBTL@EBUB ^ XSJUFSXSJUF FYBNQMF4FSJBMJ[F5P4USJOH σʔληοτʢ5'3FDPSEʣ
EFG@@EJTUPSU JNBHF NBTL@JNBHF CSJHIUOFTT@NBY@EFMUB DPOUSBTU@MPXFS DPOUSBTU@VQQFS
XJUIUGOBNF@TDPQF EJTUPSU BTTDPQF EJTUPSUFE@JNBHFUGDBTU JNBHF UGGMPBU EJTUPSUFE@JNBHFUGJNBHFSBOEPN@CSJHIUOFTT EJTUPSUFE@JNBHF NBY@EFMUBCSJHIUOFTT@NBY@EFMUB EJTUPSUFE@JNBHFUGJNBHFSBOEPN@DPOUSBTU EJTUPSUFE@JNBHF MPXFSDPOUSBTU@MPXFS VQQFSDPOUSBTU@VQQFS %BUB"VHNFOUBUJPO
%BUB"VHNFOUBUJPO ࠜઇΕ͍
XJUIUGOBNF@TDPQF SBOEPN@GMJQ@MFGU@SJHIU BTTDPQF SBOEPNUGSBOEPN@VOJGPSN EUZQFUGGMPBU EJTUPSUFE@JNBHFUGDPOE SBOEPN
MBNCEBUGJNBHFGMJQ@MFGU@SJHIU EJTUPSUFE@JNBHF MBNCEBEJTUPSUFE@JNBHF NBTL@JNBHFUGDPOE SBOEPN MBNCEBUGJNBHFGMJQ@MFGU@SJHIU NBTL@JNBHF MBNCEBNBTL@JNBHF XJUIUGOBNF@TDPQF SBOEPN@GMJQ@VQ@EPXO BTTDPQF SBOEPNUGSBOEPN@VOJGPSN EUZQFUGGMPBU EJTUPSUFE@JNBHFUGDPOE SBOEPN MBNCEBUGJNBHFGMJQ@VQ@EPXO EJTUPSUFE@JNBHF MBNCEBEJTUPSUFE@JNBHF NBTL@JNBHFUGDPOE SBOEPN MBNCEBUGJNBHFGMJQ@VQ@EPXO NBTL@JNBHF MBNCEBNBTL@JNBHF SFUVSOEJTUPSUFE@JNBHF NBTL@JNBHF
JNBHF NBTL@JNBHF@@MPBE@JNBHF GJMF@MJTU JNBHF@TJ[FJNBHF@TJ[F EJTUPSU5SVF TIVGGMF5SVF
JNBHF@CBUDI MBCFM@CBUDIUGUSBJOTIVGGMF@CBUDI <JNBHF NBTL@JNBHF> CBUDI@TJ[F'-"(4CBUDI@TJ[F DBQBDJUZDBQBDJUZ NJO@BGUFS@EFRVFVF'-"(4NJO@BGUFS@EFRVFVF OVN@UISFBET'-"(4OVN@UISFBET JNBHF@CBUDIJNBHF@CBUDI MBCFM@CBUDIMBCFM@CBUDI σʔλͷಡΈࠐΈ
EFG@@MPTT MPHJUT MBCFMT MPTTUGOOM@MPTT MPHJUTMBCFMT '-"(4CBUDI@TJ[F UGBEE@UP@DPMMFDUJPO MPTTFT MPTT
MPTTFTUGHFU@DPMMFDUJPO MPTTFT UPUBM@MPTTUGBEE@O MPTTFT OBNFUPUBM@MPTT ޡࠩ
ֶशʢ܇࿅ʣ ֶशΞϧΰϦζϜ"EBN0QUJNJ[FS ֶश ϛχόοναΠζ
ֶशʢ܇࿅ʣ
ݕূ
ࠜઇΕ͍
C-LIS CO., LTD. 4FMFDUJWF4FBSDI MCQDBTDBEF@BOJNFGBDFYNM ࠜઇΕ͍
ࠜઇΕ͍
ࠜઇΕ͍
͏·͍ͬͨ͘ɻ ͱ͜ΖͰʜʜ
݄ σΟʔϓϥʔχϯάʹΑΔࣗવݴޠॲཧͱνϟοτ#05ͷجૅ http://ascii.jp/elem/000/001/418/1418261/ ɹߨࢣɺʰ1ZUIPOͰମݧ͢ΔਂֶशʱʢίϩφࣾʣɺʰσΟʔϓϥʔχϯ άɺϏοάσʔλɺػցֶश͋Δ͍ͦͷ৺ཧֶʱʢ৽༵ࣾʣͷஶऀͰ͋Γɺ ਓೳֶձݚڀձ༏लʢ4*(-4&͜ͱֶݚڀձʮχϡʔϥϧ ωοτϫʔΫޠੜϞσϧʯ୯ஶʣΛडͷઙ৳Ұࢯɺʰ5FOTPS'MPX ͡Ί·࣮ͨ͠ફʂ࠷৽(PPHMFϚγϯϥʔχϯάʱʢΠϯϓϨε3%ɾ /FYU1VCMJTIJOHʣͷ༗ࢁܓೋࢯͰ͢ɻ
:0-03FBM5JNF0CKFDU%FUFDUJPO You only look once (YOLO) is a state-of-the-art, real-time
object detection system. On a Titan X it processes images at 40-90 FPS and has a mAP on VOC 2007 of 78.6% and a mAP of 44.0% on COCO test-dev. http://pjreddie.com/darknet/yolo/
ୈճશΞʔΩςΫνϟษڧձ ʙࣦޠΛΊ͙ͬͯʙ ɹ࣌ɿ݄ʢʣʢ։ʣ ɹॴɿגࣜձࣾϦΫϧʔτςΫϊϩδʔζΞΧσϛʔϗʔϧ ౦ژઍాؙ۠ͷάϥϯτΩϣαελϫʔ' IUUQXXXSFDSVJUKQDPNQBOZBCPVUP⒏DFIUNM ʢגࣜձࣾϦΫϧʔτςΫϊϩδʔζ༷ͷ͝ްҙʹΑΔձ͝ఏڙʣ ఆɹһɿҰൠʢࣾձਓ͓Αͼֶੜͷํʣ໊ɹֶੜ໊ ˞ఆһʹୡ࣍͠ୈకΊΒ͖ͤͯ·͢ ˞ֶੜͷํઌֶͣੜ͔Β͓ਃ͠ࠐΈԼ͍͞ɻֶੜ͕ຬ੮ͷ߹Ұൠ͔Βਃ͠ࠐΊ·͢ɻ
ࢀՃඅɿແྉ ओɹ࠵ɿ/10๏ਓશΞʔΩςΫνϟɾΠχγΞςΟϒ ޙɹԉɿגࣜձࣾυϫϯΰ https://wba-meetup.doorkeeper.jp/events/57007/
͜ͷํ๏ͷ՝
ֶशʹ͕͔͔࣌ؒΔ
ࠜઇΕ͍
ೖྗը૾͕খ͗͢͞Δ 192x192 ࠜઇΕ͍
ೖྗΛେ͖͘͢Δͱ 0VU0G.FNPSZ&SSPS
ݱࡏͷ(16 Graphic Memory: 2 GB compute capability: 5.2 (F'PSDF(59
όοναΠζΛେ͖͘͢Δͱ00.
̏νϟϯωϧʹ͢Δͱʢུ
৽͍͠(16Λߪೖ͢Δʁ (16ͷՁ֨ (16ΛΠϯετʔϧ͢Δ1$ 1$Λஔ͘ॴ Քಇ͓ͤͯ͘͞ిؾ
͘͞ΒͷߴՐྗίϯϐϡʔςΟϯά https://blog.keiji.io/2016/12/tensorflow_advent_calendar.html 「さくらの⾼高⽕火⼒力力」でTensorFlow
τϥΠΞϧ൛ʢ.BYXFMMΞʔΩςΫνϟʣ https://www.sakura.ad.jp/koukaryoku/ 9FPO$PSFʷ .FNPSZ(# 44%(#ʷ̎ 3"*% (F'PSDF(595*5"/9ʢ.BYXFMM (SBQIJD.FNPSZ(#ʣʷ̐
+-----------------------------------------------------------------------------+ | NVIDIA-SMI 367.57 Driver Version: 367.57 | |-------------------------------+----------------------+----------------------+ |
GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX TIT... Off | 0000:02:00.0 Off | N/A | | 22% 43C P2 144W / 250W | 11688MiB / 12206MiB | 88% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce GTX TIT... Off | 0000:03:00.0 Off | N/A | | 22% 35C P5 20W / 250W | 11605MiB / 12206MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 2 GeForce GTX TIT... Off | 0000:83:00.0 Off | N/A | | 22% 35C P5 19W / 250W | 11605MiB / 12206MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 3 GeForce GTX TIT... Off | 0000:84:00.0 Off | N/A | | 22% 34C P5 19W / 250W | 11603MiB / 12206MiB | 0% Default | +-------------------------------+----------------------+----------------------+ ͬͦͬͯ͘͞ΈΔ
EFG@@MPTT MPHJUT MBCFMT MPTTUGOOM@MPTT MPHJUTMBCFMT '-"(4CBUDI@TJ[F UGBEE@UP@DPMMFDUJPO MPTTFT MPTT
MPTTFTUGHFU@DPMMFDUJPO MPTTFT UPUBM@MPTTUGBEE@O MPTTFT OBNFUPUBM@MPTT MPTT@BWFSBHFTUGUSBJO&YQPOFOUJBM.PWJOH"WFSBHF OBNFBWH MPTT@BWFSBHFT@PQMPTT@BWFSBHFTBQQMZ MPTTFT <UPUBM@MPTT> XJUIUGDPOUSPM@EFQFOEFODJFT <MPTT@BWFSBHFT@PQ> UPUBM@MPTTUGJEFOUJUZ UPUBM@MPTT SFUVSOUPUBM@MPTT .VMUJ(16ରԠ https://blog.keiji.io/2016/12/tensorflow_advent_calendar.html
DJGBS@NVMUJ@HQV@USBJOQZCSFBLTXJUINPSFUIBO(16 https://github.com/tensorflow/tensorflow/issues/5652
EFG@@MPTT MPHJUT MBCFMT MPTTUGOOM@MPTT MPHJUTMBCFMT '-"(4CBUDI@TJ[F UGBEE@UP@DPMMFDUJPO MPTTFT MPTT
MPTTFTUGHFU@DPMMFDUJPO MPTTFT UPUBM@MPTTUGBEE@O MPTTFT OBNFUPUBM@MPTT IUUQTHJUIVCDPNUFOTPSGMPXUFOTPSGMPXJTTVFT SFUVSOUPUBM@MPTT .VMUJ(16ରԠ https://github.com/tensorflow/tensorflow/issues/5652
+-----------------------------------------------------------------------------+ | NVIDIA-SMI 367.57 Driver Version: 367.57 | |-------------------------------+----------------------+----------------------+ |
GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce GTX TIT... Off | 0000:02:00.0 Off | N/A | | 22% 40C P2 100W / 250W | 11688MiB / 12206MiB | 46% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce GTX TIT... Off | 0000:03:00.0 Off | N/A | | 22% 40C P2 121W / 250W | 11688MiB / 12206MiB | 35% Default | +-------------------------------+----------------------+----------------------+ | 2 GeForce GTX TIT... Off | 0000:83:00.0 Off | N/A | | 22% 38C P2 111W / 250W | 11688MiB / 12206MiB | 24% Default | +-------------------------------+----------------------+----------------------+ | 3 GeForce GTX TIT... Off | 0000:84:00.0 Off | N/A | | 22% 38C P2 132W / 250W | 11687MiB / 12206MiB | 40% Default | +-------------------------------+----------------------+----------------------+ .VMUJ(16ରԠޙͷOWJEJBTNJ
̍࣌ؒͰ̍ສεςοϓ
̐Ͱ̒ສ̓ઍεςοϓ
Oഒͷεϐʔυʂ
ೖྗը૾ΛQYʹͯ͠ 00.͕ൃੜ͠ͳ͍ʂ
࣍ճ༧ࠂ QYͷը૾Ͱֶशʢ܇࿅ʣ͠ɺإྖҬͷݕग़ਫ਼ͷมԽΛ֬ೝ͢Δɻ ϞσϧͷΈࠐΈͷϑΟϧλʔαΠζΛௐ͢Δ ʢೖྗը૾ʹରͯ͠ϑΟϧλʔ͕খ͗͢͞Δʣ ʮإྖҬݕग़ʯͱʮ؟ڸ່ͬఆʯΛಉ࣌ʹߦ͏Α͏ʹมߋ͢Δɻ
ٕज़ॻయ̎ 5FOTPS'MPXຊΛ൦͠·͢ https://goo.gl/JXXGkg ݄̐̕ʢʣ ΞΩόɾεΫΤΞ ͓ʵ
C-LIS CO., LTD. ໊֤ɾϒϥϯυ໊ɺձ໊ࣾͳͲɺҰൠʹ֤ࣾͷඪ·ͨొඪͰ͢ɻຊࢿྉதͰɺɺɺäΛׂѪ͍ͯ͠·͢ɻ ຊࢿྉɺ༗ݶձࣾγʔϦεͷஶ࡞Ͱ͢ɻܝࡌ͞Ε͍ͯΔΠϥετɺಛʹهࡌ͕ͳ͍߹ࠜઇΕ͍ͷஶ࡞Ͱ͢ɻ ຊࢿྉͷશ෦ɺ·ͨҰ෦ʹ͍ͭͯɺஶ࡞ऀ͔ΒจॻʹΑΔڐΛಘͣʹෳ͢Δ͜ͱې͡ΒΕ͍ͯ·͢ɻ