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
Firebase ML Kit for iOS Developer
Search
Kajornsak Peerapathananont
October 07, 2018
Technology
0
66
Firebase ML Kit for iOS Developer
Firebase Dev Day 2018 @Bangkok, Thailand
Kajornsak Peerapathananont
October 07, 2018
Tweet
Share
More Decks by Kajornsak Peerapathananont
See All by Kajornsak Peerapathananont
Understanding your Android build
kajornsakp
0
34
iOSDevTH #21
kajornsakp
0
40
What's new in Flutter (Google I/O Extended Bangkok 22)
kajornsakp
0
55
Mobile Design System at scale
kajornsakp
0
79
What's new in Flutter 2020
kajornsakp
0
58
Mobile Machine Learning for All Skill Levels
kajornsakp
0
30
What's new in Flutter 1.9
kajornsakp
0
49
Kotlin meets Web
kajornsakp
0
20
From design to develop with Material Components
kajornsakp
0
130
Other Decks in Technology
See All in Technology
AWSの新機能検証をやる時こそ、Amazon Qでプロンプトエンジニアリングを駆使しよう
duelist2020jp
1
250
アセスメントで紐解く、10Xのデータマネジメントの軌跡
10xinc
1
440
LiteXとオレオレCPUで作る自作SoC奮闘記
msyksphinz
0
690
Dynamic Reteaming And Self Organization
miholovesq
3
540
Amazon CloudWatch を使って NW 監視を行うには
o11yfes2023
0
170
PagerDuty×ポストモーテムで築く障害対応文化/Building a culture of incident response with PagerDuty and postmortems
aeonpeople
1
290
AIと開発者の共創: エージェント時代におけるAIフレンドリーなDevOpsの実践
bicstone
1
320
ElixirがHW化され、最新CPU/GPU/NWを過去のものとする数万倍、高速+超省電力化されたWeb/動画配信/AIが動く日
piacerex
0
150
AWSLambdaMCPServerを使ってツールとMCPサーバを分離する
tkikuchi
1
3k
Goの組織でバックエンドTypeScriptを採用してどうだったか / How was adopting backend TypeScript in a Golang company
kaminashi
6
6.1k
サーバレス、コンテナ、データベース特化型機能をご紹介。CloudWatch をもっと使いこなそう!
o11yfes2023
0
180
Automatically generating types by running tests
sinsoku
2
3.3k
Featured
See All Featured
It's Worth the Effort
3n
184
28k
Practical Orchestrator
shlominoach
186
11k
Docker and Python
trallard
44
3.3k
Being A Developer After 40
akosma
91
590k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
135
33k
Rails Girls Zürich Keynote
gr2m
94
13k
The Invisible Side of Design
smashingmag
299
50k
Bash Introduction
62gerente
611
210k
Gamification - CAS2011
davidbonilla
81
5.2k
Designing for humans not robots
tammielis
252
25k
Distributed Sagas: A Protocol for Coordinating Microservices
caitiem20
331
21k
Agile that works and the tools we love
rasmusluckow
328
21k
Transcript
ML Kit for iOS developers Kajornsak Peerapathananont Agoda
Machine Learning
#FirebaseDevDay
Google Lens
Smart Reply
On-device Machine Learning
#FirebaseDevDay Doable, but hard.
#FirebaseDevDay
#FirebaseDevDay Get Image Image Classification Transform Interpret Get Result
#FirebaseDevDay Transform unsigned char *sourceBaseAddr = (unsigned char *)(CVPixelBufferGetBaseAddress(pixelBuffer)); int
image_height; unsigned char *sourceStartAddr; if (fullHeight <= image_width) { image_height = fullHeight; sourceStartAddr = sourceBaseAddr; } else { image_height = image_width; const int marginY = ((fullHeight - image_width) / 2); sourceStartAddr = (sourceBaseAddr + (marginY * sourceRowBytes)); } const int image_channels = 4; assert(image_channels >= wanted_input_channels); tensorflow::Tensor image_tensor( tensorflow::DT_FLOAT, tensorflow::TensorShape( {1, wanted_input_height, wanted_input_width, wanted_input_channels})); auto image_tensor_mapped = image_tensor.tensor<float, 4>(); tensorflow::uint8 *in = sourceStartAddr; float *out = image_tensor_mapped.data(); for (int y = 0; y < wanted_input_height; ++y) { float *out_row = out + (y * wanted_input_width * wanted_input_channels); for (int x = 0; x < wanted_input_width; ++x) { const int in_x = (y * image_width) / wanted_input_width; const int in_y = (x * image_height) / wanted_input_height; tensorflow::uint8 *in_pixel = in + (in_y * image_width * image_channels) + (in_x * image_channels); float *out_pixel = out_row + (x * wanted_input_channels); for (int c = 0; c < wanted_input_channels; ++c) { out_pixel[c] = (in_pixel[c] - input_mean) / input_std; } } }
#FirebaseDevDay Interpret if (tf_session.get()) { std::vector<tensorflow::Tensor> outputs; tensorflow::Status run_status =
tf_session->Run( {{input_layer_name, image_tensor}}, {output_layer_name}, {}, &outputs); if (!run_status.ok()) { LOG(ERROR) << "Running model failed:" << run_status; } else { tensorflow::Tensor *output = &outputs[0]; auto predictions = output->flat<float>(); NSMutableDictionary *newValues = [NSMutableDictionary dictionary]; for (int index = 0; index < predictions.size(); index += 1) { const float predictionValue = predictions(index); if (predictionValue > 0.05f) { std::string label = labels[index % predictions.size()]; NSString *labelObject = [NSString stringWithUTF8String:label.c_str()]; NSNumber *valueObject = [NSNumber numberWithFloat:predictionValue]; [newValues setObject:valueObject forKey:labelObject]; } } dispatch_async(dispatch_get_main_queue(), ^(void) { [self setPredictionValues:newValues]; }); } }
None
#FirebaseDevDay
#FirebaseDevDay Real-world Common Use Cases
#FirebaseDevDay FIRVisionImage | VisionImage NS_SWIFT_NAME(VisionImage) @interface FIRVisionImage : NSObject @property(nonatomic,
nullable) FIRVisionImageMetadata *metadata; - (instancetype)initWithImage:(UIImage *)image NS_DESIGNATED_INITIALIZER; - (instancetype)initWithBuffer:(CMSampleBufferRef)sampleBuffer NS_DESIGNATED_INITIALIZER; - (instancetype)init NS_UNAVAILABLE; @end
Text Recognition - On-device - On-cloud
#FirebaseDevDay https://firebase.google.com/docs/ml-kit/recognize-text
#FirebaseDevDay FIRVisionText | VisionText NS_SWIFT_NAME(VisionText) @interface FIRVisionText : NSObject @property(nonatomic,
readonly) NSString *text; @property(nonatomic, readonly) NSArray<FIRVisionTextBlock *> *blocks; - (instancetype)init NS_UNAVAILABLE; @end
#FirebaseDevDay On-device Usage let textRecognizer = vision.onDeviceTextRecognizer() textRecognizer.process(visionImage) { (text,
error) in guard let text = text else { return } // do something with your text }
#FirebaseDevDay On-cloud Usage let textRecognizer = vision.cloudTextRecognize() textRecognizer.process(visionImage) { (text,
error) in guard let text = text else { return } // do something with your text }
Image Labeling - On-device (400+ labels) - On-cloud (10,000+ labels)
#FirebaseDevDay https://firebase.google.com/docs/ml-kit/label-images
#FirebaseDevDay FIRVisionLabel | VisionLabel NS_SWIFT_NAME(VisionLabel) @interface FIRVisionLabel : NSObject @property(nonatomic,
readonly) CGRect frame; @property(nonatomic, readonly) float confidence; @property(nonatomic, copy, readonly) NSString *entityID; @property(nonatomic, copy, readonly) NSString *label; @end
#FirebaseDevDay On-device Usage let labelDetector = vision.labelDetector() labelDetector.detect(in: visionImage) {
(labels, error) in guard let error == nill, let labels = labels, !labels.isEmpty else { return } // do something with your labels }
#FirebaseDevDay On-cloud Usage let labelDetector = vision.cloudLabelDetector() labelDetector.detect(in: visionImage) {
(labels, error) in guard let error == nill, let labels = labels, !labels.isEmpty else { return } // do something with your labels }
Face detection - On-device
#FirebaseDevDay
#FirebaseDevDay FIRVisionFace | VisionFace NS_SWIFT_NAME(VisionFace) @interface FIRVisionFace : NSObject @property(nonatomic,
readonly) CGRect frame; @property(nonatomic, readonly) BOOL hasTrackingID; @property(nonatomic, readonly) NSInteger trackingID; @property(nonatomic, readonly) BOOL hasHeadEulerAngleY; @property(nonatomic, readonly) CGFloat headEulerAngleY; @property(nonatomic, readonly) BOOL hasHeadEulerAngleZ; @property(nonatomic, readonly) CGFloat headEulerAngleZ; @property(nonatomic, readonly) BOOL hasSmilingProbability; @property(nonatomic, readonly) CGFloat smilingProbability; @property(nonatomic, readonly) BOOL hasLeftEyeOpenProbability; @property(nonatomic, readonly) CGFloat leftEyeOpenProbability; @property(nonatomic, readonly) BOOL hasRightEyeOpenProbability; @property(nonatomic, readonly) CGFloat rightEyeOpenProbability; - (instancetype)init NS_UNAVAILABLE; - (nullable FIRVisionFaceLandmark *)landmarkOfType:(FIRFaceLandmarkType)type; #ifdef ENABLE_FACE_CONTOUR - (nullable FIRVisionFaceContour *)contourOfType:(FIRFaceContourType)type; #endif // ENABLE_FACE_CONTOUR @end
#FirebaseDevDay On-device Usage let faceDetector = vision.faceDetector() faceDetector.detect(in: visionImage) {
(faces, error) in guard let error == nill, let faces = faces, !faces.isEmpty else { return } // do something with your faces }
#FirebaseDevDay Face Contour?
Landmark recognition - On-cloud
#FirebaseDevDay
#FirebaseDevDay FIRVisionCloudLandmark | VisionCloudLandmark NS_SWIFT_NAME(VisionCloudLandmark) @interface FIRVisionCloudLandmark : NSObject @property(nonatomic,
copy, readonly, nullable) NSString *entityId; @property(nonatomic, copy, readonly, nullable) NSString *landmark; @property(nonatomic, readonly, nullable) NSNumber *confidence; @property(nonatomic, readonly) CGRect frame; @property(nonatomic, readonly, nullable) NSArray<FIRVisionLatitudeLongitude *> *locations; - (instancetype)init NS_UNAVAILABLE; @end
#FirebaseDevDay On-cloud Usage let landmarkDetector = vision.cloudLandmarkDetector() landmarkDetector.detect(in: visionImage) {
(landmarks, error) in guard let error == nill, let landmarks = landmarks, !landmarks.isEmpty else { return } // do something with your landmarks }
Barcode scanning - On-device
#FirebaseDevDay https://firebase.google.com/docs/ml-kit/label-images
#FirebaseDevDay FIRVisionBarcode | VisionBarcode NS_SWIFT_NAME(VisionBarcode) @interface FIRVisionBarcode : NSObject @property(nonatomic,
readonly) CGRect frame; @property(nonatomic, readonly, nullable) NSString *rawValue; @property(nonatomic, readonly, nullable) NSString *displayValue; @property(nonatomic, readonly) FIRVisionBarcodeFormat format; @property(nonatomic, readonly, nullable) NSArray<NSValue *> *cornerPoints; @property(nonatomic, readonly) FIRVisionBarcodeValueType valueType; @property(nonatomic, readonly, nullable) FIRVisionBarcodeEmail *email; @property(nonatomic, readonly, nullable) FIRVisionBarcodePhone *phone; @property(nonatomic, readonly, nullable) FIRVisionBarcodeSMS *sms; @property(nonatomic, readonly, nullable) FIRVisionBarcodeURLBookmark *URL; @property(nonatomic, readonly, nullable) FIRVisionBarcodeWiFi *wifi; @property(nonatomic, readonly, nullable) FIRVisionBarcodeGeoPoint *geoPoint; @property(nonatomic, readonly, nullable) FIRVisionBarcodeContactInfo *contactInfo; @property(nonatomic, readonly, nullable) FIRVisionBarcodeCalendarEvent *calendarEvent; @property(nonatomic, readonly, nullable) FIRVisionBarcodeDriverLicense *driverLicense; - (instancetype)init NS_UNAVAILABLE; @end
#FirebaseDevDay FIRVisionBarcodeCalendarEvent | VisionBarcodeCalendarEvent NS_SWIFT_NAME(VisionBarcodeCalendarEvent) @interface FIRVisionBarcodeCalendarEvent : NSObject @property(nonatomic,
readonly, nullable) NSString *eventDescription; @property(nonatomic, readonly, nullable) NSString *location; @property(nonatomic, readonly, nullable) NSString *organizer; @property(nonatomic, readonly, nullable) NSString *status; @property(nonatomic, readonly, nullable) NSString *summary; @property(nonatomic, readonly, nullable) NSDate *start; @property(nonatomic, readonly, nullable) NSDate *end; - (instancetype)init NS_UNAVAILABLE; @end
#FirebaseDevDay On-device Usage let barcodeDetector = vision.barcodeDetector() barcodeDetector.detect(in: visionImage) {
(barcodes, error) in guard let error == nill, let barcodes = barcodes, !barcodes.isEmpty else { return } // do something with your barcodes }
Custom model - Tensorflow Lite
#FirebaseDevDay let conditions = ModelDownloadConditions(isWiFiRequired: true, canDownloadInBackground: true) let cloudModelSource
= CloudModelSource( modelName: "my_cloud_model", enableModelUpdates: true, initialConditions: conditions, updateConditions: conditions ) let registrationSuccessful = ModelManager.modelManager().register(cloudModelSource)
Demo
Thank You! #FirebaseDevDay Helpful resources fb.com/FirebaseThailand fb.com/groups/FirebaseDevTH medium.com/FirebaseThailand Kajornsak Peerapathananont