CreateMLとCoreMLでカンタンMachine Learning App開発

2018/6/21 iOSコンソーシアム Tech Working Group

Transcript

1. CreateMLͱCoreMLͰ Χϯλϯ MachineLearning App։ൃ 2018/6/21 iOSίϯιʔγΞϜTech WG WWDC2018஫໨ϙΠϯτ Mitsui Knowledge Industry ShoheiOda
2. ຊ͠Γΐ͏͸ॴͧ͘ձࣾͷެࣜͳݟ͔͍Ͱ͸ͳ͘ɺ ͍ͪ͜͡Μͷݟ͔͍Λ΋ͱʹ࡞͍͓ͤͯ͠Γ·͢ɻ ͜ͷखॻ͖෩ϑΥϯτ ׽ࣈશͥΜग़ͯ͜Ͷ͐ͳXX
3. ຊ͠Γΐ͏͸ॴͧ͘ձࣾͷެࣜͳݟ͔͍Ͱ͸ͳ͘ɺ ͍ͪ͜͡Μͷݟ͔͍Λ΋ͱʹ࡞͍͓ͤͯ͠Γ·͢ɻ Ͱ΋ Tweet΍SNS΁ͷ౤͜͏େ͔Μ͍͛ ϋογϡλά #iOSC
4. Speaker’s Profile Shohei Oda MITSUI KNOWLEDGE INDUSTRY CO,. LTD. Mobile Application Development / IoT(Device, Cloud, etc...) / UI
   UX / Web / VR / Education / etc… odasho0618 @mki_oda odasho
5. ͳʹ΍Β͍Ζ͍Ζൃද͞Εͯͨ WWDC 2018 iOS 12 SharedAR, Measure App, ARKit2 USDZ Format, Memoji, Facetime(Group Chat), Siri Shortcut, Scream Time, Smart Notifications, Shared Voice Memo, Car Play WatchOS 5 transceiver Function, Siri Shortcut Web Content, Podcast, Auto Workout, Rainbow Band tvOS 12 Support Dolby Atmos macOS Mojave(10.14) Dark Mode, CoreML2(CreateML) Desktop Stacks Movie Shooting Scream Shot Community Camera(Linkage Mac-iOS) Use UIKit for Mac Use a part of iOS App(Home, Stock) etc…
6. iOS 12 SharedAR, Measure App, ARKit2 USDZ Format, Memoji, Facetime(Group Chat), Siri Shortcut, Scream Time, Smart Notifications, Shared Voice Memo, Car Play WatchOS 5 transceiver Function, Siri Shortcut Web Content, Podcast, Auto Workout, Rainbow Band tvOS 12 Support Dolby Atmos macOS Mojave(10.14) Dark Mode, CoreML2(CreateML) Desktop Stacks Movie Shooting Scream Shot Community Camera(Linkage Mac-iOS) Use UIKit for Mac Use a part of iOS App(Home, Stock) etc… ࠓճ஫໨ͨ͠ͷ͸ίϨʜ
7. ͦ͏ͩɻDemo΍Ζ͏
8. ࠓճ͡ΎΜͼͨ͠΋ͷ n Macbook • macOS 10.14 Beta 2 (Mojave) • Xcode 10 Beta 2 n ֶ͖͔͍शνʔϜ͔Β΋ΒͬͨλϫγͱϋϦωζϛͷࣸਅ ೖख͸AppleDeveloper ͷDownloads͔Β https://developer.apple.com/download/ (AppleDeveloperͷAccount͕ͻͭΑ͏)
9. Ͱ͸࢝Ί·͠ΐ͏ɻ n Swift Playground • Get started with a playground
10. Ͱ͸࢝Ί·͠ΐ͏ɻ n Swift Playground • PlaygroundϑΝΠϧΛ࡞ͬͯίʔυΛೖྗ(̏ߦ) import CreateMLUI let builder = MLImageClassifierBuilder() builder.showInLiveView() • ࣮ߦ
11. ͳΜ͔ग़͖ͯͨɻ
12. ֶशDataΛDrug & Drop ͱͦͷલʹ…
13. ༻ҙ͢ΔֶशDataʹ͍ͭͯ Χϯλϯʹߟ͑ͯΈΔɻ n Training Data • ֶ͖͔͍शͤ͞Δσʔλ Θ͔Δ n Evaluation Data • ֶ͖͔͍शͨ͠σʔλΛͻΐ͏͔͢Δσʔλ ·͊Θ͔Δ n Validation Data • ֶ͖͔͍शͨ͠σʔλΛνϡʔχϯά͢ΔͨΊͷσʔλ Θ͔Βͳ͍
14. ΑΓΧϯλϯʹߟ͑ͯΈΔ Training Data Validation Data Evaluation Data ͓ษڧ ΋͚͗͠Μ ຊ൪͚͠Μ ͚͠ΜΛ;ΜͩΜʹ;͘ΜͰ͓Γ·͢
15. Training Data ʜ Folder͜͏͍ͤ1 Hari Tawashi
16. Training Data Folder͜͏͍ͤ2 Hari.1.jpg Tawashi.1.jpg Hari.2.jpg ʜ
17. ͖ͭͮ͸DemoͰ
18. Appendix n Swift Playground • CodeͷΈͰmlmodelͷ࡞͍ͤ΋Ͱ͖Δ import Foundation import CreateML // Specify Data let trainDirectory = URL(fileURLWithPath: "/Users/shoheioda/Desktop/Machine_Learning/TAG-IMG/Train_Vali/") let testDirectory = URL(fileURLWithPath: "/Users/shoheioda/Desktop/Machine_Learning/TAG-IMG/EvaluationData/")
19. Appendix n Swift Playground • CodeͷΈͰmlmodelͷ࡞͍ͤ΋Ͱ͖Δ // Create Model let model = try MLImageClassifier(trainingData: .labeledDirectories(at: trainDirectory)) // Evaluate Model let evaluation = model.evaluation(on: .labeledDirectories(at: testDirectory)) // Save Model try model.write(to: URL(fileURLWithPath: "/Users/shoheioda/Desktop/Hari_TawashiClassifier.mlmodel"))
20. Appendix n Sample App • DemoͰ࢖༻ͨ͠Sample App͸ɺ͍͔ͷURL͔ΒDLͰ͖Δɻ https://developer.apple.com/documentation/vision/classifying_images_with_vision_and_ core_ml