Lecuture on Deep Learning API
?Download as PPTX, PDF?
0 likes?169 views
For deep learning engineer and server engineer, this slides illustrates how to create VGG image recognition API with Flask, Keras in Python, which is very simple and easy to get started. All code shown in slides can be found in gist https://gist.github.com/asterisk37n/e139272963f1bd659fcd532a35c59978 This slides is used in a lecture done on June 1, at Shibuya.
![20
from flask import Flask, jsonify
app = Flask(__name__)
@app.route('/users')
@app.route('/users/<int:id>')
def users(id=None):
print(id)
users = [
{'id': 1, 'name': 'Alice'},
{'id': 2, 'name': 'Bob'},
{'id': 3, 'name': 'Charlie'}]
if id is None:
return jsonify(users)
elif isinstance(id, int) and 1 <= id <= 3:
return jsonify({'id': id, 'name': users[id-1]['name']})
else:
return jsonify({})
if __name__ == '__main__':
app.run()](https://image.slidesharecdn.com/hldbed1lszmjyhlv642z-signature-afde9ee73ab19e9680d3be8ebf9bc3643442a7b0a892aaeb7b7469b51272a252-poli-180605072802/85/Lecuture-on-Deep-Learning-API-20-320.jpg)
![Response
{
"success":true,
"predictions":[
{
"label":"animal",
"probability":0.7
},
{
"label":"human",
"probability":0.2
},
{
"label":"fish",
"probability":0.1
}
]
}
これから作るVGG API RequestとResponseの定義
Request
{"file":"binary image data"}
51](https://image.slidesharecdn.com/hldbed1lszmjyhlv642z-signature-afde9ee73ab19e9680d3be8ebf9bc3643442a7b0a892aaeb7b7469b51272a252-poli-180605072802/85/Lecuture-on-Deep-Learning-API-51-320.jpg)
![画像を認識する
@app.route('/', methods=['GET', 'POST'])
def upload_file():
response = {'success': False}
if request.method == 'POST':
if request.files.get('file'):
img_requested = request.files['file'].read()
img = Image.open(io.BytesIO(img_requested))
if img.mode != 'RGB':
img = img.convert('RGB')
img = img.resize((224, 224))
img = image.img_to_array(img)
img = np.expand_dims(img, axis=0)
inputs = preprocess_input(img)
preds = model.predict(inputs)
results = decode_predictions(preds) 54](https://image.slidesharecdn.com/hldbed1lszmjyhlv642z-signature-afde9ee73ab19e9680d3be8ebf9bc3643442a7b0a892aaeb7b7469b51272a252-poli-180605072802/85/Lecuture-on-Deep-Learning-API-54-320.jpg)
![VGGの出力結果をjsonの形に整形する
response['predictions'] = []
for (imagenetID, label, prob) in results[0]:
row = {'label': label, 'probability': float(prob)}
response['predictions'].append(row)
response['success'] = True
return jsonify(response)
55](https://image.slidesharecdn.com/hldbed1lszmjyhlv642z-signature-afde9ee73ab19e9680d3be8ebf9bc3643442a7b0a892aaeb7b7469b51272a252-poli-180605072802/85/Lecuture-on-Deep-Learning-API-55-320.jpg)
Lecuture on Deep Learning API
- 1. Deep Learning API By Keras and Flask On June 1, 2018 at Shibuya supported by CoLab WATANABE Naoki
- 2. 自己紹介 2
- 3. 渡邊直樹 2017- スタートアップ創業など 2016-2017 IBMで研究開発 2010-2016 東大でCERNの素粒子実験開発 ● スキューバのライセンスホルダー ● シドニーマラソン2017完走 ● 富士下山90分 ● 40以上の国を訪問 ● 400+冊の読書 (2017) ● 自然語学の勉強が趣味 Colabではtinder自動操作についてのLT経験あり。 3
- 4. 目次 1. API 1. APIが何かを知る 2. Flaskで画像アップローダーを作る 2. Deep Learning 1. Deep Learningが何かを知る 2. VGGで画像認識を行う 3. Deep Learningを使ったAPIを作る 4
- 5. pipインストールを前提とするもの ● Flask ● Tensorflow ● Keras ● PIllow *Python 3.6.1 5
- 7. 1.APIを理解して Flaskで実装するWhat is API and How to make it?
- 12. わずか7行のapp.pyを書きpython app.pyで実行 from flask import Flask app = Flask(__name__) @app.route('/') def hello_world(): return 'Hello, World!' if __name__ == '__main__': app.run() 12 URL html contents start server
- 13. Go to http://localhost:5000, and see 13
- 14. エンドポイントを追加する 14
- 15. 15 from flask import Flask app = Flask(__name__) @app.route('/') def index(): return 'Index Page' @app.route('/hello') def hello(): return 'Hello, World' @app.route('/users/<username>') def show_user_profile(username): # show the user profile for that user return 'User %s' % username if __name__ == '__main__': app.run()
- 16. 16
- 17. 17
- 18. 18
- 19. REST APIぽくする 19
- 20. 20 from flask import Flask, jsonify app = Flask(__name__) @app.route('/users') @app.route('/users/<int:id>') def users(id=None): print(id) users = [ {'id': 1, 'name': 'Alice'}, {'id': 2, 'name': 'Bob'}, {'id': 3, 'name': 'Charlie'}] if id is None: return jsonify(users) elif isinstance(id, int) and 1 <= id <= 3: return jsonify({'id': id, 'name': users[id-1]['name']}) else: return jsonify({}) if __name__ == '__main__': app.run()
- 21. 21
- 22. 22
- 23. 23
- 25. 25
- 26. 26
- 27. 27
- 28. 28
- 29. 2. Deep learningの 概念を理解する。 Kerasを使う。 How deep learning works and how to use it
- 31. 31
- 36. 36
- 37. 全結合層 37
- 40. ディープラーニングには学習が必要 1. データの用意 2. ネットワークの学習 3. ネットワークの精度を確認 40
- 41. ● GoogLeNet ● Xception ● ResNet50 ● InceptionV3 ● VGG16 ● VGG19 など。 ただし、学習済みのネットワークが公開されている 41
- 42. Kerasでは一部の学習済みモデルが簡単に利用可能 from keras.applications.xception import Xception from keras.applications.vgg16 import VGG16 from keras.applications.vgg19 import VGG19 from keras.applications.resnet50 import ResNet50 from keras.applications.inception_v3 import InceptionV3 model = VGG16(weights='imagenet', include_top=True) 42
- 43. VGG16 ● ImageNet Large Scale Visual Recognition Challengeが初出 (2014) ● 13層の畳み込み層 ● 3層のフル結合層 ● 1000種類の画像を認識 ● 知られたモデルの基礎にVGG16が使われている(写真: keras document)。 43
- 46. まず、VGGモデルの概要を表示(vgg_sample.py) from keras.applications.vgg16 import VGG16 model = VGG16(weights='imagenet', include_top=True) print(model.summary()) 46
- 47. 47
- 48. 画像を認識 (vgg_sample.py) from keras.applications.vgg16 import VGG16, preprocess_input, decode_predictions from keras.preprocessing import image import numpy as np img_path = 'cat.jpg' model = VGG16(weights='imagenet', include_top=True) img = image.load_img(img_path, target_size=(224, 224)) x = image.img_to_array(img) x = np.expand_dims(x, axis=0) x = preprocess_input(x) pred = model.predict(x) label = decode_predictions(pred) print(label) 48
- 49. Result (正解！) ● Tiger cat (42%) ● Egyptian cat (36%) ● Tabby (12%) ● Lynx (2%) 出力結果は、(imagenetID, ラベル, 確率)のリスト。 49
- 52. import os from flask import Flask, request, redirect, url_for, jsonify, Response from werkzeug.utils import secure_filename from keras.applications.vgg16 import VGG16, preprocess_input, decode_predictions from keras.preprocessing import image import numpy as np from PIL import Image import io モジュールのimportとモデルの読み込みを構築。 52
- 53. appの宣言とモデルの読み込みを構築。 53 app = Flask(__name__) model = None def load_model(): global model model = VGG16(weights='imagenet', include_top=True)
- 54. 画像を認識する @app.route('/', methods=['GET', 'POST']) def upload_file(): response = {'success': False} if request.method == 'POST': if request.files.get('file'): img_requested = request.files['file'].read() img = Image.open(io.BytesIO(img_requested)) if img.mode != 'RGB': img = img.convert('RGB') img = img.resize((224, 224)) img = image.img_to_array(img) img = np.expand_dims(img, axis=0) inputs = preprocess_input(img) preds = model.predict(inputs) results = decode_predictions(preds) 54
- 55. VGGの出力結果をjsonの形に整形する response['predictions'] = [] for (imagenetID, label, prob) in results[0]: row = {'label': label, 'probability': float(prob)} response['predictions'].append(row) response['success'] = True return jsonify(response) 55
- 56. オプションで、画像のアップロード画面 return ''' <!doctype html> <title>Upload new File</title> <h1>Upload new File</h1> <form method=post enctype=multipart/form-data> <p><input type=file name=file> <input type=submit value=Upload> </form> ''' 56
- 57. モデルを読み込み、アプリケーションを実行 (load_modelだけでmodelを読み込むようにしている) if __name__ == '__main__': load_model() # About threaded=False: https://github.com/keras- team/keras/issues/2397#issuecomment-377914683 # avoid model.predict runs before model initiated app.run(threaded=False) 57
- 61. 61
- 62. 参考文献 Flask Quickstart http://flask.pocoo.org/docs/1.0/quickstart/ Building a simple Keras + deep learning REST API https://blog.keras.io/building-a-simple-keras-deep-learning-rest-api.html Imagenet 1000 classes and labels https://gist.github.com/aaronpolhamus/964a4411c0906315deb9f4a3723aac57 62
- 63. That’s all, folks. Thank you. 63