Tweet
Tweet

Slide 1

Slide 1 text

For Android devs

Slide 2

Slide 2 text

Attila Blénesi Senior Android Engineer @ Halcyon Mobile @ablenessy

Slide 3

Slide 3 text

Is machine learning only for experts ? @ablenessy | @droidconRO | #DroidconT

Slide 4

Slide 4 text

No content

Slide 5

Slide 5 text

Application Developer ML Practitioner Data Scientist Firebase ML Kit Machine Learning APIs TensorFlow Cloud ML @ablenessy | @droidconRO | #DroidconT

Slide 6

Slide 6 text

Firebase ML Kit @ablenessy | @droidconRO | #DroidconT Source: firebase.google.com/products/ml-kit

Slide 7

Slide 7 text

Text recognition

Slide 8

Slide 8 text

val image = FirebaseVisionImage.fromBitmap(selectedImage) val detector = FirebaseVision.getInstance() .getVisionTextDetector() detector.detectInImage(image) .addOnSuccessListener { texts -> processTextRecognitionResult(texts) }a .addOnFailureListener(...) @ablenessy | @droidconRO | #DroidconT

Slide 9

Slide 9 text

Text recognition

Slide 10

Slide 10 text

Text recognition

Slide 11

Slide 11 text

val image = FirebaseVisionImage.fromBitmap(selectedImage) val options = FirebaseVisionCloudDetectorOptions.Builder() .setModelType(FirebaseVisionCloudDetectorOptions.LATEST_MODEL) .setMaxResults(15) .build() val detector = FirebaseVision.getInstance() .getVisionCloudDocumentTextDetector(options) detector.detectInImage(image) .addOnSuccessListener { texts -> processTextRecognitionResult(texts) }a .addOnFailureListener(...) @ablenessy | @droidconRO | #DroidconT

Slide 12

Slide 12 text

Application Developer ML Practitioner Data Scientist Firebase ML Kit Machine Learning APIs TensorFlow Cloud ML

Slide 13

Slide 13 text

Machine Learning APIs

Slide 14

Slide 14 text

Source: cloud.google.com/vision/

Slide 15

Slide 15 text

google.com/about/stories/machine-learning-qa/

Slide 16

Slide 16 text

google.com/about/stories/machine-learning-qa/

Slide 17

Slide 17 text

Source: cloud.google.com/vision/

Slide 18

Slide 18 text

Source: cloud.google.com/vision/

Slide 19

Slide 19 text

Do I even have to dig deeper ? @ablenessy | @droidconRO | #DroidconT

Slide 20

Slide 20 text

generic model = generic solution Do I even have to dig deeper ? @ablenessy | @droidconRO | #DroidconT

Slide 21

Slide 21 text

Application Developer ML Practitioner Data Scientist Firebase ML Kit Machine Learning APIs TensorFlow Cloud ML

Slide 22

Slide 22 text

TensorFlow Cloud ML #edges2cats

Slide 23

Slide 23 text

TensorFlow Cloud ML #edges2cats

Slide 24

Slide 24 text

TensorFlow Cloud ML #edges2cats Source: https://affinelayer.com/pixsrv/

Slide 25

Slide 25 text

workflow Model > Training > Optimisation > Deployment @ablenessy | @droidconRO | #DroidconT

Slide 26

Slide 26 text

How does pix2pix work? Model > Training > Optimisation > Deployment

Slide 27

Slide 27 text

How does pix2pix work? Model > Training > Optimisation > Deployment

Slide 28

Slide 28 text

How does pix2pix work?

Slide 29

Slide 29 text

How does pix2pix work?

Slide 30

Slide 30 text

How does pix2pix work? generator INPUT OUTPUT

Slide 31

Slide 31 text

How does pix2pix work? generator INPUT OUTPUT That doesn’t look right

Slide 32

Slide 32 text

How does pix2pix work?

Slide 33

Slide 33 text

Training the discriminator

Slide 34

Slide 34 text

Training the generator

Slide 35

Slide 35 text

Model > Training > Optimisation > Deployment

Slide 36

Slide 36 text

Model > Training > Optimisation > Deployment

Slide 37

Slide 37 text

Model > Training > Optimisation > Deployment

Slide 38

Slide 38 text

Model > Training > Optimisation > Deployment

Slide 39

Slide 39 text

Local ~ 36 hours ~ 12 hours Cloud vs @ablenessy | @droidconRO | #DroidconT

Slide 40

Slide 40 text

Model > Training > Optimisation > Deployment Cloud ML

Slide 41

Slide 41 text

Model > Training > Optimisation > Deployment gcloud ml-engine jobs submit training $JOB_NAME \ --job-dir $JOB_DIR \ --module-name trainer.pix2pix \ --package-path ./trainer \ --region $REGION \ --config=trainer/cloudml-gpu.yaml \ -- \ --mode train \ --input-dir gs://$BUCKET_NAME/train Cloud ML

Slide 42

Slide 42 text

Model > Training > Optimisation > Deployment gcloud ml-engine jobs submit training $JOB_NAME \ --job-dir $JOB_DIR \ --module-name trainer.pix2pix \ --package-path ./trainer \ --region $REGION \ --config=trainer/cloudml-gpu.yaml \ -- \ --mode train \ --input-dir gs://$BUCKET_NAME/train Cloud ML

Slide 43

Slide 43 text

trainingInput: scaleTier: CUSTOM masterType: standard_gpu # 1 GPU pythonVersion: "3.5" runtimeVersion: "1.8" Model > Training > Optimisation > Deployment Cloud ML

Slide 44

Slide 44 text

trainingInput: scaleTier: CUSTOM masterType: complex_model_m_gpu # 4 GPUs pythonVersion: "3.5" runtimeVersion: "1.8" Model > Training > Optimisation > Deployment Cloud ML

Slide 45

Slide 45 text

Model > Training > Optimisation > Deployment gcloud ml-engine jobs submit training $JOB_NAME \ --job-dir $JOB_DIR \ --module-name trainer.pix2pix \ --package-path ./trainer \ --region $REGION \ --config=trainer/cloudml-gpu.yaml \ -- \ --mode train \ --input-dir gs://$BUCKET_NAME/train Cloud ML

Slide 46

Slide 46 text

Model > Training > Optimisation > Deployment gcloud ml-engine jobs submit training $JOB_NAME \ --job-dir $JOB_DIR \ --module-name trainer.pix2pix \ --package-path ./trainer \ --region $REGION \ --config=trainer/cloudml-gpu.yaml \ -- \ --mode train \ --input-dir gs://$BUCKET_NAME/train Cloud ML

Slide 47

Slide 47 text

Model > Training > Optimisation > Deployment

Slide 48

Slide 48 text

Model > Training > Optimisation > Deployment TensorBoard

Slide 49

Slide 49 text

Model > Training > Optimisation > Deployment TensorBoard

Slide 50

Slide 50 text

Model > Training > Optimisation > Deployment TensorBoard tensorboard —logdir /path/to/model http://localhost:6006/

Slide 51

Slide 51 text

Model > Training > Optimisation > Deployment TensorBoard

Slide 52

Slide 52 text

Model > Training > Optimisation > Deployment

Slide 53

Slide 53 text

Model > Training > Optimisation > Deployment ~680 mb ~2600 operations

Slide 54

Slide 54 text

Model > Training > Optimisation > Deployment ~680 mb ~2600 operations

Slide 55

Slide 55 text

Model > Training > Optimisation > Deployment ~200 mb ~400 operations

Slide 56

Slide 56 text

Model > Training > Optimisation > Deployment checkpoint reduced_model.data-00000-of-00001 reduced_model.meta reduced_model.index frozen_model.pb

Slide 57

Slide 57 text

Model > Training > Optimisation > Deployment # We start a session using a temporary fresh Graph with tf.Session(graph=tf.Graph()) as sess: # We import the meta graph in the current default Graph saver = tf.train.import_meta_graph(input_checkpoint + ‘.meta’,…) saver.restore(sess, input_checkpoint) # We restore the weights output_graph_def = tf.graph_util.convert_variables_to_constants( sess, # The session is used to retrieve the weights tf.get_default_graph().as_graph_def(), # retrieve the nodes output_node_names # select the useful nodes ) with tf.gfile.GFile(output_graph, "wb") as f: f.write(output_graph_def.SerializeToString())

Slide 58

Slide 58 text

Model > Training > Optimisation > Deployment checkpoint reduced_model.data-00000-of-00001 reduced_model.meta reduced_model.index frozen_model.pb

Slide 59

Slide 59 text

Model > Training > Optimisation > Deployment // TensorFlow Android implementation ‘org.tensorflow:tensorflow-android:1.10.0’

Slide 60

Slide 60 text

Model > Training > Optimisation > Deployment

Slide 61

Slide 61 text

Model > Training > Optimisation > Deployment output_tensor Generated image jalammar.github.io/Supercharging-android-apps-using-tensorflow/ Generator

Slide 62

Slide 62 text

Model > Training > Optimisation > Deployment val interpreter = TensorFlowInferenceInterface( assetManager, "file:///android_asset/frozen_model.pb" ) inputValues = generateInput(inputDrawing) interpreter.feed("input_node", inputValues, inputDimensions) interpreter.run(arrayOf("output_node"), false) interpreter.fetch(“output_node", outputValues) val result = generateBitmapFromOutput(outputValues)

Slide 63

Slide 63 text

No content

Slide 64

Slide 64 text

No content

Slide 65

Slide 65 text

Model > Training > Optimisation > Deployment // TensorFlow Lite implementation ‘org.tensorflow:tensorflow-lite:1.10.0’

Slide 66

Slide 66 text

Model > Training > Optimisation > Deployment val interpreter = Interpreter( File(URI.create("file://android_asset/frozen_model.tflite")) )a inputValues = generateInput(inputDrawing) interpreter.run(inputValues, outputValues) val result = generateBitmapFromOutput(outputValues)

Slide 67

Slide 67 text

Demo @ablenessy | @droidconRO | #DroidconT

Slide 68

Slide 68 text

@ablenessy | @droidconRO | #DroidconT Machine learning is NOT just for experts

Slide 69

Slide 69 text

@ablenessy | @droidconRO | #DroidconT Machine learning is NOT just for experts

Slide 70

Slide 70 text

@ablenessy | @droidconRO | #DroidconT