TensorFlow no Android – v2

Palestra apresentada no DevFest Campinas

Transcript

1. IA no bolso com TensorFlow e Android Douglas Drumond @douglasdrumond
2. @douglasdrumond Por que IA no telefone? 2
3. @douglasdrumond O que é IA? Separar bolinhas 3
4. @douglasdrumond O que é IA? 4
5. @douglasdrumond O que é IA? Ham vs Spam 5
6. @douglasdrumond Ham vs Spam Gramaticalmente correto? Oferece viagra? Dear sir/madam? Anexos? 6
7. @douglasdrumond DeepBlue 7
8. @douglasdrumond 8
9. @douglasdrumond Go (ғޟ) Impossível 9
10. @douglasdrumond DeepMind 10
11. @douglasdrumond AlphaGo 11
12. @douglasdrumond Vamos aprender IA! 12
13. @douglasdrumond Como eu me sinto Axiomas: • Existe 1 • Existe uma operação s(n) que obtém o sucessor Prove: • Todo número par maior que 2 é soma de dois primos 13
14. @douglasdrumond Coisas de IA Redes Neurais? Modelos?
 Perceptrons? SVM? CNN? WTF? 14
15. @douglasdrumond TensorFlow 15
16. @douglasdrumond O que é? “An open-source software library for Machine Intelligence” – https://www.tensorflow.org/ 16
17. @douglasdrumond Arquiteturas Android iOS Raspberry Pi 17
18. @douglasdrumond Instale Bazel: https://bazel.build/versions/master/docs/install.html Instale TensorFlow: https://www.tensorflow.org/install/install_sources 18
19. @douglasdrumond Usando TensorFlow no Android Studio Configurar bazel no build.gradle: 19 def bazelLocation = '/usr/local/bin/bazel'
20. @douglasdrumond Task do bazel task buildNativeBazel(type: Exec) { workingDir '../../..' commandLine bazelLocation, 'build', '-c', 'opt', \ 'tensorflow/examples/android:tensorflow_native_libs', \ '--crosstool_top=//external:android/crosstool', \ '--cpu=' + cpuType, \ '--host_crosstool_top=@bazel_tools//tools/cpp:toolchain' } 20 Não se preocupe em entender agora, pode copiar dos exemplos
21. @douglasdrumond Opção 2 Baixar do repositório 21
22. @douglasdrumond Diferenças entre esses métodos? Tamanho do APK 22
23. @douglasdrumond NDK Configurar WORKSPACE 23 # Uncomment and update the paths in these entries to build the Android demo. android_sdk_repository( name = "androidsdk", api_level = 23, # Ensure that you have the build_tools_version below installed in the # SDK manager as it updates periodically. build_tools_version = "26.0.1", # Replace with path to Android SDK on your system path = "<PATH_TO_SDK>", )
24. @douglasdrumond NDK Pelo menos versão 14 24 android_ndk_repository( name="androidndk", path="<PATH_TO_NDK>", # This needs to be 14 or higher to compile TensorFlow. # Please specify API level to >= 21 to build for 64-bit # architectures or the Android NDK will automatically select biggest # API level that it supports without notice. # Note that the NDK version is not the API level. api_level=14)
25. @douglasdrumond Live Demo 25
26. @douglasdrumond Inspiração 26
27. @douglasdrumond Live Demo 27
28. @douglasdrumond Linguagens TensorFlow: C++ Apps: Java ou Kotlin WAT? 28
29. @douglasdrumond Android Inference Library Prebuilt binaries at https://ci.tensorflow.org/view/Nightly/job/ nightly-android/ Inference docs at https://github.com/tensorflow/tensorflow/ blob/master/tensorflow/contrib/android/java/org/tensorflow/ contrib/android/TensorFlowInferenceInterface.java Full TensorFlow Java API bridges with C++
 https://github.com/tensorflow/tensorflow/tree/master/java
 29
30. @douglasdrumond Antes do Android Treine seu modelo 30 import tensorflow as tf from tensorflow.examples.tutorials.mnist import input_data ... optimizer = tf.train.AdamOptimizer( learning_rate=learning_rate).minimize(cost)
31. @douglasdrumond Gere o protobuf from tensorflow.python.tools import freeze_graph ... freeze_graph.freeze_graph(input_graph_path, input_saver_def_path, input_binary, input_checkpoint_path, output_node_names, restore_op_name, filename_tensor_name, output_graph_path, clear_devices) 31
32. @douglasdrumond Como funciona TensorFlowImageClassifier carrega • MODEL file • LABEL file • TensorFlowInferenceInterface (uma classe que carrega a lib nativa) 32
33. @douglasdrumond Como funciona? List<Classifier.Recognition> results = classifier.recognizeImage(bitmap); …
 inferenceInterface.feed(inputName, floatValues, 1, inputSize, inputSize, 3); …
 inferenceInterface.run(outputNames, logStats); … inferenceInterface.fetch(outputName, outputs); 33
34. @douglasdrumond Outras preocupações Tamanho do modelo •Inception v3 = 90 MB 34
35. @douglasdrumond Use as ferramentas do TF freeze_graph Graph Transformation Tool •strip_unused_nodes •remove_nodes •fold_batch_norms 35 Quantize weights 75% de melhora no tamanho Memory mapping
36. @douglasdrumond Outras preocupações Tamanho do modelo •Inception v3 = 90 MB •Inception v3 quantized = 24 MB •Inception v1 quantized = 7MB 36
37. @douglasdrumond E o apk? Aumento de 12MB antes de tuning com Inception v3 37
38. @douglasdrumond Tuning Inclua apenas as operações necessárias Rode python/tools/ print_selective_registration_header.py 
 Ponha ops_to_register.h na raiz do TensorFlow
 Compile com -DSELECTIVE_REGISTRATION 38
39. @douglasdrumond E o apk? Aumento de 2MB depois do tuning 39
40. @douglasdrumond Ou seja… O que estão esperando? 40
41. @douglasdrumond Onde aprender? • https://codelabs.developers.google.com/codelabs/ tensorflow-for-poets/ • https://github.com/tensorflow/tensorflow/ • https://omid.al/posts/2017-02-20-Tutorial-Build-Your- First-Tensorflow-Android-App.html • https://www.tensorflow.org • Udacity 41
42. @douglasdrumond Mais fontes https://medium.com/joytunes/slimming-down- tensorflows-impact-on-your-apk-size-9b5cb07e6238 https://medium.com/joytunes/deploying-a-tensorflow- model-to-android-69d04d1b0cba 42
43. Grazie mille! @douglasdrumond