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
Wantedly Peopleのスキャン画面の裏側 / Wantedly People Sca...
Search
Sponsored
·
Ship Features Fearlessly
Turn features on and off without deploys. Used by thousands of Ruby developers.
→
Shinichi Goto
May 31, 2017
Technology
6.6k
6
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
Wantedly Peopleのスキャン画面の裏側 / Wantedly People Scanning Screen
2017/05/31 Wantedly技術見学会 iOS編
Shinichi Goto
May 31, 2017
More Decks by Shinichi Goto
See All by Shinichi Goto
WWDC18 ML Overview
_shingt
1
1.5k
Core ML 🏃 iOS Engineer
_shingt
1
670
Core ML / Vision Frameworkを使ってできること / What can we achieve using Core ML and Vision framework
_shingt
3
4.5k
Providing Better Feedback in Real-time Object Detection Apps
_shingt
2
1.6k
Value Types in WWDC16
_shingt
3
2.6k
debug-remote-local-notification-on-watchos
_shingt
0
11k
Server Side Swift question
_shingt
3
1.2k
Other Decks in Technology
See All in Technology
なぜ Platform Engineering の土台に Kubernetes を選ぶのか
r4ynode
1
450
チームで進めるAI駆動アジャイル×ウォーターフォール
kumaiu
0
140
AI駆動開発が変える、大規模開発の前提 ーHuman in the Loop から Human on the Loop へ / AIE2026
visional_engineering_and_design
30
23k
あなたの AI ワークスペースに、 専門コーダーを連れてくる - Amazon Quick Desktop 最新情報
kawaji_scratch
1
120
AI-DLCを活用した高品質・安全なAI駆動開発実践 / AI Driven Development with AI-DLC
yoshidashingo
0
160
Ruby::Boxでできること、Refinementsでできること
joker1007
3
410
"何を作るか"を任される エンジニアは、どう育つのか
yutaokafuji
1
410
実装は速くなった、レビューはどうする? ― 自身のレビューをAIで再現させるサーヴァントエンジニアリングのすゝめ / Implementation got faster. So what about reviews? — An invitation to Servant Engineering: Recreating your own code reviews with AI
nrslib
8
4.4k
AIを「創る」と「使う」の循環 — HRテックが実践するリアルなAI組織実装
taketo957
0
1.8k
新規事業を牽引する技術選定 〜フルスタックTypeScript開発の実践事例〜
nullnull
3
380
GoとSIMDとWasmの今。
askua
3
520
社内 AI エージェント Synapse と セマンティックレイヤーの育て方
hiroakis
1
1.3k
Featured
See All Featured
Have SEOs Ruined the Internet? - User Awareness of SEO in 2025
akashhashmi
0
370
Agile Leadership in an Agile Organization
kimpetersen
PRO
0
160
Why Mistakes Are the Best Teachers: Turning Failure into a Pathway for Growth
auna
0
150
Gemini Prompt Engineering: Practical Techniques for Tangible AI Outcomes
mfonobong
2
430
B2B Lead Gen: Tactics, Traps & Triumph
marketingsoph
0
140
Speed Design
sergeychernyshev
33
1.8k
VelocityConf: Rendering Performance Case Studies
addyosmani
333
25k
Efficient Content Optimization with Google Search Console & Apps Script
katarinadahlin
PRO
1
600
Max Prin - Stacking Signals: How International SEO Comes Together (And Falls Apart)
techseoconnect
PRO
0
180
Lightning talk: Run Django tests with GitHub Actions
sabderemane
0
200
Designing for Performance
lara
611
70k
How Software Deployment tools have changed in the past 20 years
geshan
0
34k
Transcript
Wantedly Peopleͷ εΩϟϯը໘ͷཪଆ ɹ 2017/05/31 Wantedlyٕज़ݟֶձ iOSฤ shingt (Shinichi Goto)
@ Wantedly, Inc.
shingt (Shinichi Goto) GitHub: @shingt Twi5er: @_shingt
None
None
Outline • εΩϟϯը໘ͷॲཧͷྲྀΕ • ٕज़ཁૉ / ߏ • ෆ҆ఆͳݕग़݁Ռͷରॲ •
σόοά൛ͷ
εΩϟϯը໘ͷॲཧͷྲྀΕ
None
None
None
None
None
None
None
None
None
None
ٕज़ཁૉ / ߏ
• AVFounda*on • OpenCV • ίϯϐϡʔλϏδϣϯ༻ͷϥΠϒϥϦ • ໊ྖҬͷݕग़ɺͦͷଞը૾ॲཧʹར༻ • ಛผͳͷ͍ͬͯͳ͍
໊ݕग़ͷOpenCVҎ֎ͷબࢶ • CoreImage • Accerelate (BNNS) • Metal (CNN) •
TensorFlow
ݕग़ϩδοΫAndroidͱڞ௨Խ͍ͨ͠ ʢগͳ͘ͱॳظϦϦʔεͰʣ
=> Accerelate / Metal / CoreImage ݟૹΓ ʢલऀ2ͭiOS9Ͱར༻Ͱ͖ͳ͍ʣ
TensorFlow • ΞϓϦʹΈࠐΈɺผ్ֶशࡁͷϞσϧΛద༻ɺͱ͍͏ͱ͜Ζ ·Ͱࢼͨ͠ • ໊ݕग़෦ͷ࣮ࡍͷ׆༻ʹࢸΒͣ • ։ൃظؒɺσʔληοτɺਫ਼ͷ
༨ஊ: Google I/O '17 pod 'TensorFlow-experimental' ɹ • Android Meets
TensorFlow: How to Accelerate Your App with AI • "experimental" Ͱͳ͍ํ "Coming soon" ͱͳͬͯΔ • h?ps:/ /cocoapods.org/pods/TensorFlow
ߏ
ɹ • OpenCV • C++Ͱهड़͞ΕͨΠϯλʔϑΣΠεΛར༻ • => Swi/͔Βͷͷݺͼग़͠ෆՄ • ʢඞવతʹʣݕग़ؔ࿈ͷϩδοΫObjec3ve-C++ଆʹدͤΔ
None
// CardDetector.mm - (NSArray<CardFeature *> *)cardFeaturesInMat:(const cv::Mat &)mat { std::vector<dtct::Card>
cards = ... // Detect cards on C++ layer NSMutableArray<CardFeature *> *cardFeatures = [NSMutableArray array]; for (const auto& card : cards) { if (!card.is_namecard) continue; CardFeature *cardFeature = [[CardFeature alloc] initWithDTCTCard:card]; [cardFeatures addObject:cardFeature]; } return cardFeatures; }
CardDetector • C++ͷϨΠϠͷProxyʹ͍ۙҐஔ • ࣮ࡍʹ͋Δఔը૾ॲཧϩδοΫؚΉ • ࠲ඪิਖ਼ɺಁࢹมɺetc. • σόΠεଆͰಈ͔ͯ͠ΈͯॳΊͯݟ͔ͭΔɺͱ͍͏͋ͬ ͨͨΊ
C++ͷϨΠϠ • Γग़ͯ͠։ൃ & ୯ಠͰϏϧυ • ΞϓϦશମΛϏϧυ͢Δͷʹൺͯѹతʹ͍ͨΊ • ݕग़෦ͷϩδοΫΛվળ͍͢͠ʢͬͱجຊతͳ෦ PythonͰ࣮ݧʣ
• దٓiOSϓϩδΣΫτऔΓࠐΈ
ෆ҆ఆͳݕग़݁Ռͷରॲ
໊ྖҬͷݕग़
AVCaptureStillImageOutput ʢ੩ࢭը༻ʣ OR AVCaptureVideoDataOutputʢಈը༻ʣ
໊εΩϟϯʹγϟολʔԻ໐Βͨ͘͠ͳ͍ => AVCaptureVideoDataOutput
: iPhone༳ΕΔ
=> ϒϨΔ
None
ղܾࡦ: γʔέϯεͱͯ͠ѻ͏
None
None
None
None
ݕग़݁Ռͷථ • ϑϨʔϜͷ໌ʢͲΕ͚ͩϒϨ͍ͯͳ͍͔ʣ • ϥϓϥγΞϯϑΟϧλΛ͔͚ͨ݁ՌͷࢄʢͰ͋ΔఔݟΕ Δʣ • ݕग़໊ຕ • ͲΕ͚ͩۙͷϑϨʔϜ͔
ࢀߟ: ໌ͷԋࢉ @implementation UIImage (CVExtension) - (double)sharpness { cv::Mat mat,
gray, laplacian; UIImageToMat(self, mat); cv::cvtColor(mat, gray, CV_BGR2GRAY); cv::Laplacian(gray, laplacian, CV_64F); cv::Scalar mean, stddev; cv::meanStdDev(laplacian, mean, stddev, cv::Mat()); return stddev.val[0] * stddev.val[0]; } @end
None
None
τϥοΩϯά࣌
None
None
None
ɹ • ݁ߏ҆ͳղܾࡦͰ͋Δ • ͱ͍͑ܶతʹମݧ্ !
σόοά൛ͷ
ϦΞϧλΠϜͳը૾ॲཧΞϓϦͷ։ൃͰࠔΔ͜ͱ • ൃੜ࣌ɺݱঢ়Ѳʹ͕͔͔࣌ؒΓ͍͢ • ϓϩηεͷͲͷ෦͕ѱ͍ͷ͔ • ઢݕग़ʁ࠲ඪมʁ໌ͷԋࢉʁ
Debug, QAͰ • ԋࢉঢ়گɺϦιʔεঢ়گΛදࣔ • ໌, γʔέϯεॲཧ, CPU, etc. •
ը૾ॲཧͷաఔΛදࣔ
None
None
None
ɹ • ࢹ֮తʹ͔Γ͍ͨ͢ΊΛݟ͚͍ͭ͢ • ͍͚ͦ͏ײ͕CEOʹΘΔ
·ͱΊ • εΩϟϯը໘ͷॲཧͷྲྀΕ / ٕज़ཁૉ / ߏ Λհ • ෆ҆ఆͳݕग़݁Ռͷରॲ
=> γʔέϯεຖʹॲཧ • "ΊͪΌͪ͘ΌεϚʔτʂ" ͳ͜ͱͯ͠ͳ͍ʢݱঢ়ʣ • Debug/QA൛ͰॲཧաఔɺϦιʔεঢ়گΛग़͓ͯ͘͠ • ΓޙBlogͰެ։ʢ͢Δ͔ʣ
Thanks!