Tensor Flow for Android Devs

Slide 1

Slide 1 text

@ablenessy | @droidconLisbon | #DCLISBON19 for android devs

Slide 2

Slide 2 text

@ablenessy | @droidconLisbon | #DCLISBON19 What can AI do?

Slide 3

Slide 3 text

@ablenessy | @droidconLisbon | #DCLISBON19 What can AI do? Image classiﬁcation Object detection Gesture recognition Speech recognition Text generation Translate Generate music Generate images

Slide 4

Slide 4 text

@ablenessy | @droidconLisbon | #DCLISBON19 Artiﬁcial Intelligence Machine Learning Deep learning: Computer vision, NLP, GAN …

Slide 5

Slide 5 text

@ablenessy | @droidconLisbon | #DCLISBON19 Source: tensorﬂow.com/about

Slide 6

Slide 6 text

@ablenessy | @droidconLisbon | #DCLISBON19 Source: tensorﬂow.com/about Neural Network Neural Network

Slide 7

Slide 7 text

@ablenessy | @droidconLisbon | #DCLISBON19 Source: tensorﬂow.com/about Training

Slide 8

Slide 8 text

@ablenessy | @droidconLisbon | #DCLISBON19 Caﬀe PyTorch

Slide 9

Slide 9 text

@ablenessy | @droidconLisbon | #DCLISBON19 TensorFlow https://www.tensorﬂow.org/beta You can write code in: C++, Python, Swift, JavaScript Deploy to: CPU, GPU, TPU Mobile iOS and Android Raspberry Py, Coral (edge TPU), Microcontrollers TensorFlow 2.0 is now in beta Premade Estimators tf.Keras tf.* Low level API High level API Ready made models

Slide 10

Slide 10 text

@ablenessy | @droidconLisbon | #DCLISBON19 Keras High level Neural Networks API, written in Python Integrated in TensorFlow : tf.keras Sequential - best to start with Functional - more ﬂexible Model subclassing - extend a Model class Deep learning with Python By Francois Chollet

Slide 11

Slide 11 text

@ablenessy | @droidconLisbon | #DCLISBON19 import tensorflow as tf mnist = tf.keras.datasets.mnist (x_train, y_train),(x_test, y_test) = mnist.load_data() x_train, x_test = x_train / 255.0, x_test / 255.0 model = tf.keras.models.Sequential([ tf.keras.layers.Flatten(input_shape=(28, 28)), tf.keras.layers.Dense(512, activation=tf.nn.relu), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10, activation=tf.nn.softmax) ]) model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) model.fit(x_train, y_train, epochs=5) model.evaluate(x_test, y_test) Digit classiﬁer

Slide 12

Slide 12 text

@ablenessy | @droidconLisbon | #DCLISBON19 On device ML Low latency, no server calls Works oﬄine, no connection needed Privacy, data stays on device

Slide 13

Slide 13 text

@ablenessy | @droidconLisbon | #DCLISBON19 Tensor Flow Open sourced Tensor Flow Mobile TF Lite Developer preview ML Kit Tensor Flow 2.0 TF Lite - ﬁnal TF Mobile deprecated 2015 2016 2017 2018 2019 Source: https://speakerdeck.com/margaretmz

Slide 14

Slide 14 text

@ablenessy | @droidconLisbon | #DCLISBON19 Application Developer ML Practitioner Data Scientist Firebase ML Kit TensorFlow Lite Neural Networks API

Slide 15

Slide 15 text

@ablenessy | @droidconLisbon | #DCLISBON19 Firebase ML Kit Source: ﬁrebase.google.com/products/ml-kit Image labelling Text Recognition Face Detection Barcode scanning Landmark detection Supports Custom models Dynamic downloads New: Auto ML, Smart reply, Object detection, Translation

Slide 16

Slide 16 text

@ablenessy | @droidconLisbon | #DCLISBON19 Text Recognition

Slide 17

Slide 17 text

@ablenessy | @droidconLisbon | #DCLISBON19 val image = FirebaseVisionImage.fromBitmap(selectedImage) val detector = FirebaseVision.getInstance() .getVisionTextDetector() detector.detectInImage(image) .addOnSuccessListener { texts -> processTextRecognitionResult(texts) } .addOnFailureListener(...)

Slide 18

Slide 18 text

@ablenessy | @droidconLisbon | #DCLISBON19 ML workﬂow Get Data > Train > Convert > Integrate

Slide 19

Slide 19 text

@ablenessy | @droidconLisbon | #DCLISBON19 TensorFlow #pix2pix Source: https://aﬃnelayer.com/pixsrv/

Slide 20

Slide 20 text

@ablenessy | @droidconLisbon | #DCLISBON19 #pix2pix TensorFlow

Slide 21

Slide 21 text

@ablenessy | @droidconLisbon | #DCLISBON19 #pix2pix TensorFlow

Slide 22

Slide 22 text

@ablenessy | @droidconLisbon | #DCLISBON19 How does pix2pix work? Source: https://aﬃnelayer.com/pix2pix/

Slide 23

Slide 23 text

@ablenessy | @droidconLisbon | #DCLISBON19 How does pix2pix work? Source: https://aﬃnelayer.com/pix2pix/

Slide 24

Slide 24 text

@ablenessy | @droidconLisbon | #DCLISBON19 Source: https://dzone.com/articles/working-principles-of-generative-adversarial-netwo

Slide 25

Slide 25 text

@ablenessy | @droidconLisbon | #DCLISBON19 How does pix2pix work? generator INPUT OUTPUT Source: https://aﬃnelayer.com/pix2pix/

Slide 26

Slide 26 text

@ablenessy | @droidconLisbon | #DCLISBON19 How does pix2pix work? Source: https://aﬃnelayer.com/pix2pix/

Slide 27

Slide 27 text

@ablenessy | @droidconLisbon | #DCLISBON19

Slide 28

Slide 28 text

@ablenessy | @droidconLisbon | #DCLISBON19

Slide 29

Slide 29 text

Source: https://aﬃnelayer.com/pix2pix/

Slide 30

Slide 30 text

@ablenessy | @droidconLisbon | #DCLISBON19 TensorBoard

Slide 31

Slide 31 text

@ablenessy | @droidconLisbon | #DCLISBON19 Train > Convert > Integrate

Slide 32

Slide 32 text

~680 mb ~2600 operations Train > Convert > Integrate

Slide 33

Slide 33 text

~680 mb ~2600 operations Train > Convert > Integrate

Slide 34

Slide 34 text

~200 mb ~400 operations Train > Convert > Integrate

Slide 35

Slide 35 text

@ablenessy | @droidconLisbon | #DCLISBON19 Train > Convert > Integrate

Slide 36

Slide 36 text

@ablenessy | @droidconLisbon | #DCLISBON19 Interpreter Core Converter TensorFlow Lite Format Operation Kernels Hardware acceleration ~75 KB vs 1.1 MB in TF ~2.4 MB vs 22.3 MB in Deprecate TF Mobile

Slide 37

Slide 37 text

@ablenessy | @droidconLisbon | #DCLISBON19 Converter TensorFlow Lite Format   Command line Python API Android App (Java /C++ API) iOS App (C++ API) Linux (e.g. Raspberry Pi) (C++ API) Trained TensorFlow Model SavedModel HDF5

Slide 38

Slide 38 text

Convert saved_model_dir = PATH + 'saved_model' tf.saved_model.save(generator, saved_model_dir) converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir) tf_lite_model = converter.convert() tf_lite_model_file.write_bytes(tf_lite_model)

Slide 39

Slide 39 text

Convert NEW saved_model_dir = PATH + 'saved_model' tf.saved_model.save(generator, saved_model_dir) converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir) tf_lite_model = converter.convert() tf_lite_model_file.write_bytes(tf_lite_model) converter.optimizations = [tf.lite.Optimize.DEFAULT] converter.target_spec.supported_types = [tf.float16] tf_lite_fp16_model = converter.convert() tf_lite_f16_model_file.write_bytes(tf_lite_fp16_model)

Slide 40

Slide 40 text

Optimisation options 103 MB 207 MB 154 MB 61 MB - requires custom ops Fails

Slide 41

Slide 41 text

Test test test!

Slide 42

Slide 42 text

Model > Training > Deployment android { aaptOptions { noCompress "tflite" } } implementation 'org.tensorflow:tensorflow-lite:1.14.0'

Slide 43

Slide 43 text

@ablenessy | @droidconLisbon | #DCLISBON19 Model > Training > Deployment

Slide 44

Slide 44 text

@ablenessy | @droidconLisbon | #DCLISBON19 Attila Blénesi Android Engineer @ Babylon Health @ablenessy @ablenessy ablenesi