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
google/mediapipe で始めるARアプリ開発/iOSDC2020
Search
noppefoxwolf
September 21, 2020
Programming
1.7k
1
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
google/mediapipe で始めるARアプリ開発/iOSDC2020
noppefoxwolf
September 21, 2020
More Decks by noppefoxwolf
See All by noppefoxwolf
High performance GIF playback/iOSDC25
noppefoxwolf
1
620
Spice up your notifications/try!Swift25
noppefoxwolf
3
840
iOSの隠されたAPIを解明し、開発効率を向上させる方法/iOSDC24
noppefoxwolf
2
1.3k
今から理解するApp Intentエコシステム/WWDC24Recap
noppefoxwolf
0
74
既存アプリをvisionOS対応してリリースした話/visionOS LT vol5
noppefoxwolf
0
230
UIのブラックボックスを探る/iOSDC23
noppefoxwolf
3
5.1k
CoreGraphicsでドット絵を描こう/iOSDC22
noppefoxwolf
0
3.2k
ランタイムデバッグのススメ/iOSDC21
noppefoxwolf
1
5.7k
モバイルファーストなアプリを作るためにvearがしたこと/xRDCC
noppefoxwolf
0
200
Other Decks in Programming
See All in Programming
そこに3びきプロダクトがいるじゃろう——生成AI時代における“価値が届かない理由”の構造
kosuket
0
430
FDEが実現するAI駆動経営の現在地
gonta
2
270
Go 1.27 における memory allocation の高速化
andpad
0
110
為什麼你並不需要ViewModel / No, you don't need a ViewModel
lovee
1
480
ソフトウェア設計に溶けるインフラ ― AWS CDK のインフラ認識論
konokenj
3
760
Detecting Compromised CI with eBPF and Cilium Tetragon
lizrice
0
160
ドリフトを絶対に許さない(?)CDK運用 / CDK Ops with Zero Tolerance for Drifts (?)
akihisaikeda
1
180
【やさしく解説 設計編・中級 #6】良いアーキテクチャとは ~ 一本の登り道の、行き先 ~
panda728
PRO
0
210
仕様駆動開発の消費期限
watany
16
7.2k
「人を評価する AI」の設計と実装
ryoyanara
0
180
Apache Hive: そしてCloud Native Lakehouseへ
okumin
1
210
生成AIで帳票OCRが「簡単に」作れる時代になった?
kon_shou
0
620
Featured
See All Featured
No one is an island. Learnings from fostering a developers community.
thoeni
21
3.8k
The AI Search Optimization Roadmap by Aleyda Solis
aleyda
1
6k
Bioeconomy Workshop: Dr. Julius Ecuru, Opportunities for a Bioeconomy in West Africa
akademiya2063
PRO
1
210
Designing for Performance
lara
611
70k
For a Future-Friendly Web
brad_frost
183
10k
The SEO identity crisis: Don't let AI make you average
varn
0
530
Between Models and Reality
mayunak
4
390
What’s in a name? Adding method to the madness
productmarketing
PRO
24
4.1k
Statistics for Hackers
jakevdp
799
230k
Navigating Weather and Climate Data
rabernat
0
460
The #1 spot is gone: here's how to win anyway
tamaranovitovic
3
1.1k
The Limits of Empathy - UXLibs8
cassininazir
1
600
Transcript
google/mediapipe Ͱ࢝ ΊΔARΞϓϦ։ൃ ϨΪϡϥʔτʔΫʢ20ʣ noppe 1
ର • ARΞϓϦʹڵຯ͕͋Δ • τϥοΩϯάʹڵຯ͕͋Δ • ػցֶश͔Βͳͯ͋͘Γ·ͤΜɻ 2
noppe • גࣜձࣾσΟʔɾΤψɾΤʔ • ιʔγϟϧϥΠϒΞϓϦ Pococha • ݸਓ։ൃऀ • vear
ReplayKitͰεϚϗ͔Β৴Ͱ͖Δ VTuberΞϓϦ • iOSDC18,19 • ͖ͭͶ͕͖ 3
4
5
noppe • גࣜձࣾσΟʔɾΤψɾΤʔ • ιʔγϟϧϥΠϒΞϓϦ Pococha • ݸਓ։ൃऀ • vear
ReplayKitͰεϚϗ͔Β৴Ͱ͖Δ VTuberΞϓϦ • iOSDC18,19 • ͖ͭͶ͕͖ 6
mediapipe ͱʁ 7
mediapipe github.com/google/mediapipe • ॲཧ(Culculator)Λܨ͛ͯύΠϓϥΠϯʢGraphʣΛߏங͢Δ πʔϧ • ओʹը૾ॲཧͷύΠϓϥΠϯߏஙʹΘΕΔ • MLॲཧΛࠞͥࠐΉࣄ͕ग़དྷΔ 8
9
10
mediapipe Culculatorɺ࣮ࡁΈͷ͑Δ • LuminanceCalculator RGBΛड͚औΓɺًը૾Λग़ྗ͢Δ • SobelEdgesCalculator SobelϑΟϧλΛ͔͚ͨը૾Λग़ྗ͢Δ ...etc 11
12
࣮ࡁΈͷGraph ͜ΕΒͷ࣮͙͢ʹ͑Δios ࠓ͜ΕΛ͏ྲྀΕΛղઆ ios Ұ෦iOSͰ࣮͕ͳ͍ͷ͋Δ 13
ΞδΣϯμ • άϥϑͷϏϧυ • ARKitͱͷ࿈ܞ • Ͱ͖ΔࣄɾͰ͖ͳ͍ࣄ 14
άϥϑͷϏϧυ 15
github.com/noppefoxwolf/HandTracker ࠓճ͏ιʔείʔυ 16
Graph • ߏஙͨ͠GraphόΠφϦͱͯ͠Ϗϧυ ͢Δ͜ͱ͕Ͱ͖Δ • όΠφϦΛObjC͔Β͏࣮͕͋Δͷ ͰɺͦΕΒΛͬͯϑϨʔϜϫʔΫΛ࡞ Δࣄ͕Ͱ͖Δɻ • ࠓσϞͰհͨ͠ϋϯυτϥοΧʔ
ΛϑϨʔϜϫʔΫͱͯ͠Ϗϧυ 17
18
άϥϑͷϏϧυ ϏϧυπʔϧΛͬͯϏϧυΛߦ͏ • bazelbuild/bazel (ϕΠθϧ) • KubernetesTensorFlowͰ࠾༻ Track E -
Ryo Aoyama BazelΛར༻ͨ͠Micro Modular Architecture 19
άϥϑͷϏϧυ bazelͷΠϯετʔϧεΫϦϓτམͱ͖࣮ͯͯ͠ߦ͢Δ͚ͩ curl -LO "https://.../bazel-3.2.0-installer-darwin-x86_64.sh" chmod +x "bazel-3.2.0-installer-darwin-x86_64.sh" ./bazel-3.2.0-installer-darwin-x86_64.sh 20
άϥϑͷϏϧυ XcodeͰϏϧυ͕Ͱ͖Δtulsi - bazelbuild/tulsi tulsi BuildPhaseScriptͰbazelΛୟ͚ͩ͘ͳͷͰɺඞਢͰͳ͍ 21
άϥϑͷϏϧυ tulsiϏϧυεΫϦϓτୟ͚ͩ͘ͰΠϯετʔϧͰ͖Δʢཁbazelʣ git clone
[email protected]
:bazelbuild/tulsi.git ./build_and_run.sh 22
ͦͷ΄͔ඞཁͳͷ • Xcode • ࣌ؒ • ωοτϫʔΫ 23
ϏϧυͷྲྀΕ 1. ϑϨʔϜϫʔΫͷίʔυΛॻ͘ 2. BUILDϑΝΠϧʹՌͷใґଘੑͳͲΛهड़͢Δ 3. bazelͰbuild͢Δ 4. framework͕ग़དྷ্͕Δ 24
ϑϨʔϜϫʔΫͷίʔυ Λॻ͘ 25
ϑϨʔϜϫʔΫͷίʔυΛॻ͘ 1. binarypbΛಡΈࠐΜͰMPPGraphΛ࡞Δ 2. MPPGraphΛ։࢝͢Δ 3. MPPGraphʹը૾ΛૹΔ 4. delegateͰ݁ՌΛड͚औΔ 26
27
binarypbΛಡΈࠐΜͰMPPGraphΛ࡞ Δ + (instanceType) init { NSURL* url = [NSURL
...@"hand_landmarks.binarypb"]; NSData* data = [NSData dataWithContentsOfURL:url options:0 error:nil]; mediapipe::CalculatorGraphConfig config; config.ParseFromArray(data.bytes, data.length); self.graph = [[MPPGraph alloc] initWithGraphConfig:config]; } 28
MPPGraphΛ։࢝͢Δ GraphͰར༻͢Δ֤छίϯϙʔωϯτͷॳظԽ͕ߦΘΕΔ - (void)start { [graph startWithError: nil]; } 29
MPPGraphʹը૾ΛૹΔ - (void)sendPixelBuffer:(CVPixelBufferRef)pixelBuffer { [self.mediapipeGraph sendPixelBuffer:pixelBuffer intoStream:@"input_video" packetType:MPPPacketTypePixelBuffer]; } 30
delegateͰ݁ՌΛड͚औΔ - (void)mediapipeGraph:(MPPGraph*)graph didOutputPacket:(const ::mediapipe::Packet&)packet fromStream:(const std::string&)streamName { const auto&
timestamp = packet.Timestamp().Value(); const auto& landmarks = packet.Get<::mediapipe::NormalizedLandmarkList>(); NSArray *landmarkObjects = [self toObject: landmarks]; [delegate didOutputLandmarks: landmarkObjects]; } 31
BUILDϑΝΠϧॻ͘ 32
BUILDϑΝΠϧͱ bazelͰϏϧυ͢ΔϑΝΠϧґଘؔΛهड़ͨ͠ϑΝΠϧ 33
objc_library( name = "HandTrackerLibrary", hdrs = ["HandTracker.h"], srcs = ["HandTracker.mm"],
data = [ "hand_tracking:hand_tracking_mobile_gpu_binary_graph", ... ], deps = [ "//mediapipe/objc:mediapipe_framework_ios", "//mediapipe/objc:mediapipe_input_sources_ios", "//mediapipe/graphs/hand_tracking:mobile_calculators", "//mediapipe/framework/formats:landmark_cc_proto", ], ), ) 34
35
bazelͰϏϧυ͢Δ $ bazel build mediapipe/iosdc:HandTracker 36
framework͕ग़དྷ্͕Δ githubʹஔ͍͓͖ͯ·ͨ͠ https://github.com/noppefoxwolf/HandTracker 37
ARKitͱͷ࿈ܞ 38
ARKitͱͷ࿈ܞ ARKitͷ্༷ग़དྷͳ͍ࣄΛmediapipeͰิ͏͜ͱͰɺදݱྗΛղ ์Ͱ͖Δ • AR্ۭؒͷΦϒδΣΫτΛखͰૢ࡞͢Δ • ݕग़ͨ͠ΦϒδΣΫτͷԣʹΩϟϥΫλʔΛஔ͢Δ • إͷࠎ֨Λิਖ਼͠ͳ͕Β৭Λม͑Δhair hair
Hair Segmentationݱঢ়iOSඇରԠͰ͕͢ 39
ARKitͱͷ࿈ܞΞϓϦ։ൃ ͜ΜͳͷΛ࡞Γ·͢ 40
41
ARKit࿈ܞͷྲྀΕ 1. ARFrame͔ΒcaptureImageΛऔΓग़͢ 2. PixelFormatΛม 3. TrackerૹΔ 4. DelegateͰ݁ՌΛड͚औΔ 5.
RealityKitͷϘλϯΛ্ԼɾΧϯτΛ දࣔ 42
ARFrame͔ΒcaptureImageΛऔΓग़͢ • captureImage = Χϝϥ͔Βͷө૾ arSession.delegate = self ... //
ARSessionDelegate func session(_ session: ARSession, didUpdate frame: ARFrame) { let captureImage: CVPixelBuffer = frame.capturedImage } 43
PixelFormatΛม ը૾σʔλ͕ͲͷΑ͏ʹ֨ೲ͞Ε͍ͯΔ͔Λද͍ͯ͠Δ • ARKitͰYCbCrܗࣜ • mediapipeBGRAͷΈΛड͚औΕΔ →มͷඞཁ͕͋Δ 44
kCVPixelFormatType_32BGRA 1ϐΫηϧ32bitͰදݱ͢ΔϑΥʔϚοτ B:FF G:00 R:00 A:FF → 45
kCVPixelFormatType_420YpCbCr8BiPl anarFullRange • 2ຕͷY(8bit)+CbCr(16bit)ͷΈ߹ΘͤͰ1ϑϨʔϜΛදݱ͢Δ ϑΥʔϚοτ • ܰྔͳͷͰϏσΦσʔλͰΘΕΔ͜ͱ͕ଟ͍ 46
YCbCr͔ΒBGRAͷม • Accelerate • vImage • Metal • MSL GPUΛ͏Metalͷํ͕ߴ
47
YCbCr͔ΒBGRAͷม ࣍ͷܭࢉࣜͰม͕Ͱ͖Δ R = Y + 1.402 × Cr G
= Y - 0.344136 × Cb - 0.714136 × Cr B = Y + 1.772 × Cb A = 1 48
BlueDress YCbCr͔ΒBGRAʹߴίϯόʔτ͢ΔϥΠϒϥϦ https://github.com/noppefoxwolf/BlueDress 49
TrackerૹΔ let captureImage = frame.capturedImage let bgraCaptureImage = converter.convertToBGRA(captureImage) handTracker.send(bgraCaptureImage)
50
DelegateͰ݁ՌΛड͚औΔ ؔઅͷҐஔX,Y,ZͰऔΕ·͢…͕ɺੈք࠲ඪܥͰແ͘ X,Y: Screen࠲ඪܥ Z: खट͔ΒͷԞߦ͖ ͔͠औΕ·ͤΜɻ೦ʂ 51
52
RealityKitͷϘλϯΛ্ԼɾΧϯτΛ දࣔ ࠓճਓࠩ͠ࢦͷԆઢ্ʹEntity͕͋ΕɺisPressΛtrueʹ͢Δ func handTracker(_ handTracker: HandTracker!, didOutputLandmarks landmarks: [Landmark]!)
{ let indexFinderPosition = landmarks[8] let screenLocation: CGPoint = ...(indexFinderPosition) self.isPress = arView .entities(at: screenLocation) .contains(where: { $0.id == self.buttonEntity.id }) } 53
RealityKitͷϘλϯΛ্ԼɾΧϯτΛ දࣔ private func press() { self.scene.notifications.press.post() count += 1
changeText("\(count)") } 54
55
56
Ͱ͖ΔࣄͰ͖ͳ͍ࣄ 57
ؾʹͳΔͱ͜Ζ • HandTrackerΧϝϥ͔Βͷڑ͕औΕΔΘ͚Ͱͳ͍ࣄʹҙ →IrisͰԞߦ͖͕औΕΔ • ֤ؔઅͷճసऔΕͳ͍ →ࣗͰܭࢉͯࣗ͠વʹݟͤΔ͕ඞཁ • ϑϨʔϜϫʔΫαΠζ͕·͊·͊Ͱ͔͍ →100MB͘Β͍
58
͍͍ͱ͜Ζ • ࣮ࡁΈͷάϥϑ͕ඇৗʹ࣭͕ߴ͍ • OSS • ϚϧνϓϥοτϑΥʔϜ 59
ͬͯΈ͍ͨ͜ͱ • HairSegmentationͳͲΛiOSʹରԠ͍ͯ͠ͳ͍ͷԿނͳͷ͔ௐ ͍ͨ • NSObjectΛܦ༝͠ͳ͚ΕߴԽ͕ΊΔͷͰ • ϥϯυϚʔΫ͚ͩΛग़ྗ͢ΔάϥϑΛ࡞ͬͯΈ͍ͨ 60
͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝ ·ͨ͠ 61