DjangoでPOST処理時に他のファイル実行させ、結果を受け取る方法

Djangoで

python manege.py runserver

でアプリ(myapp)を起動します。ホーム画面(index.html)は画像をフォームにアップロードし、それが特定のディレクトリ（'/Users/downloads/django_app/myapp/image/myapp'）に保存される形式になっています。（アプリの形式の参考サイト：https://qiita.com/narupo/items/e3dbdd5d030952d10661、
https://qiita.com/okoppe8/items/86776b8df566a4513e96）

特定のディレクトリから画像を読み込んで、予測結果を返すCNNのコードがmain.pyにあります。

自分がやりたいこと：
フォームに画像を投稿した時、POST時の処理としてmain.pyを起動させ、その変数を受け取り、index.htmlにレンダリングする。
・質問
views.pyのPOST時の処理のとき、main.pyを実行させる方法はどうやるのでしょうか？

個人的にはmain.pyの予測結果を変数（LABEL）に格納して、import文main.pyを呼び出して、レンダリングすることを考えています

python
1# views.py
2
3from django.shortcuts import render, redirect
4from .forms import PhotoForm
5from .models import Photo
6
7def index(req):
8    if req.method == 'GET':
9        return render(req, 'myapp/index.html', {
10                      'form': PhotoForm(),
11                      })
12    
13    if req.method == 'POST':
14        ==POST時の処理==

python
1# index.html
2<form action="{% url 'index' %}" method="POST" enctype="multipart/form-data">
3    {% csrf_token %}
4    {{ form }}
5    <input type="submit" value="投稿" />
6</form>

python
1#main.py
2
3from __future__ import print_function
4from __future__ import absolute_import
5
6import warnings
7import os
8import numpy as np
9import tensorflow as tf
10from keras.optimizers import SGD
11from keras import layers
12from keras.preprocessing import image
13from keras.applications.imagenet_utils import decode_predictions
14from keras.models import Model
15from keras.layers import Activation,AveragePooling2D, BatchNormalization, Concatenate, Conv2D,Dense
16from keras.layers import GlobalAveragePooling2D, GlobalMaxPooling2D, Input,Lambda, MaxPooling2D
17from keras import backend as K
18from keras import metrics
19from keras.preprocessing import image
20import matplotlib.pyplot as plt
21from keras.callbacks import EarlyStopping
22tf.logging.set_verbosity(tf.logging.ERROR)
23
24
25def InceptionResNetV2(img_input,include_top=True,
26                      pooling=None, classes=13):
27
28    # Stem block: 35 x 35 x 192
29    x = conv2d_bn(img_input, 32, 3, strides=2, padding='valid')
30    x = conv2d_bn(x, 32, 3, padding='valid')
31    x = conv2d_bn(x, 64, 3)
32    x = MaxPooling2D(3, strides=2)(x)
33    x = conv2d_bn(x, 80, 1, padding='valid')
34    x = conv2d_bn(x, 192, 3, padding='valid')
35    x = MaxPooling2D(3, strides=2)(x)
36
37    # Mixed 5b (Inception-A block): 35 x 35 x 320
38    branch_0 = conv2d_bn(x, 96, 1)
39    branch_1 = conv2d_bn(x, 48, 1)
40    branch_1 = conv2d_bn(branch_1, 64, 5)
41    branch_2 = conv2d_bn(x, 64, 1)
42    branch_2 = conv2d_bn(branch_2, 96, 3)
43    branch_2 = conv2d_bn(branch_2, 96, 3)
44    branch_pool = AveragePooling2D(3, strides=1, padding='same')(x)
45    branch_pool = conv2d_bn(branch_pool, 64, 1)
46    branches = [branch_0, branch_1, branch_2, branch_pool]
47    channel_axis = 1 if K.image_data_format() == 'channels_first' else 3
48    x = Concatenate(axis=channel_axis, name='mixed_5b')(branches)
49
50    # 10x block35 (Inception-ResNet-A block): 35 x 35 x 320
51    for block_idx in range(1, 11):
52        x = inception_resnet_block(x,
53                                   scale=0.17,
54                                   block_type='block35',
55                                   block_idx=block_idx)
56
57    # Mixed 6a (Reduction-A block): 17 x 17 x 1088
58    branch_0 = conv2d_bn(x, 384, 3, strides=2, padding='valid')
59    branch_1 = conv2d_bn(x, 256, 1)
60    branch_1 = conv2d_bn(branch_1, 256, 3)
61    branch_1 = conv2d_bn(branch_1, 384, 3, strides=2, padding='valid')
62    branch_pool = MaxPooling2D(3, strides=2, padding='valid')(x)
63    branches = [branch_0, branch_1, branch_pool]
64    x = Concatenate(axis=channel_axis, name='mixed_6a')(branches)
65
66    # 20x block17 (Inception-ResNet-B block): 17 x 17 x 1088
67    for block_idx in range(1, 21):
68        x = inception_resnet_block(x,
69                                   scale=0.1,
70                                   block_type='block17',
71                                   block_idx=block_idx)
72
73    # Mixed 7a (Reduction-B block): 8 x 8 x 2080
74    branch_0 = conv2d_bn(x, 256, 1)
75    branch_0 = conv2d_bn(branch_0, 384, 3, strides=2, padding='valid')
76    branch_1 = conv2d_bn(x, 256, 1)
77    branch_1 = conv2d_bn(branch_1, 288, 3, strides=2, padding='valid')
78    branch_2 = conv2d_bn(x, 256, 1)
79    branch_2 = conv2d_bn(branch_2, 288, 3)
80    branch_2 = conv2d_bn(branch_2, 320, 3, strides=2, padding='valid')
81    branch_pool = MaxPooling2D(3, strides=2, padding='valid')(x)
82    branches = [branch_0, branch_1, branch_2, branch_pool]
83    x = Concatenate(axis=channel_axis, name='mixed_7a')(branches)
84
85    # 10x block8 (Inception-ResNet-C block): 8 x 8 x 2080
86    for block_idx in range(1, 10):
87        x = inception_resnet_block(x,
88                                   scale=0.2,
89                                   block_type='block8',
90                                   block_idx=block_idx)
91    x = inception_resnet_block(x,
92                               scale=1.,
93                               activation=None,
94                               block_type='block8',
95                               block_idx=10)
96
97    # Final convolution block: 8 x 8 x 1536
98    x = conv2d_bn(x, 1536, 1, name='conv_7b')
99
100    # Classification block
101    x = GlobalAveragePooling2D(name='avg_pool')(x)
102    return Dense(classes, activation='softmax', name='predictions')(x)
103
104input_ = Input(batch_shape=(None,100,100,3))
105output_ = InceptionResNetV2(img_input=input_)
106model = Model(input_, output_, name='inception_resnet_v2')
107model.load_weights('/Users/downloads/Downloads/django_app/myapp/weght_dir/model.h5')
108model.compile(optimizer=SGD(decay=0.1, momentum=0.9, nesterov=True),
109                            loss='categorical_crossentropy',
110                        metrics=['accuracy'])
111img_path = '/Users/downloads/django_app/myapp/image/myapp/images.jpeg'
112img = image.load_img(img_path, target_size=(100, 100))
113x = image.img_to_array(img)
114x = np.expand_dims(x, axis=0)
115x = preprocess_input(x)
116preds = model.predict(x)
117print('Predicted:', decode_predictions(preds))

行動規範の内容に同意します

回答1件

ベストアンサー

Githubに似たような構成のプロジェクトのサンプルプロジェクトが有ったような気がしますが。

1, CNNの学習済みモデルを事前に保存しておく。
Keras modelを保存するには？
2, main.pyをモジュール化して
2-1, 上記学習済みモデルを呼び出すコードを記述。
2-2, 画像のpreprocessとmodel.predict(x)を行い結果を返す関数を作成し。views.py側で呼び出す。

投稿2018/09/02 14:36