flaskとOpenCVによる画像認識のエラー

前提・実現したいこと

flaskにて、OpenCVを活用した
簡単な画像認識アプリを作成してます。
（読み込んだ画像の分類）

読み込んだ画像の分類結果を返すresult.htmlに
遷移する段階にて、以下のようなエラーが発生してしまい、よく意味がわからず、
非常に困っております。

具体的な原因、解決策がお分かりの方、
ぜひご教示頂ければ幸いです。

発生している問題・エラーメッセージ

raise ValueError("Tensor %s is not an element of this graph." % obj)

pyファイル

import pickle
import numpy as np
from flask import Flask, render_template, request, redirect, url_for

from keras.models import load_model
import cv2
import numpy as np
from keras.models import load_model
import matplotlib.pyplot as plt
import io
from keras.preprocessing.image import img_to_array, load_img

app = Flask(__name__)

# 機械学習モデルの読込
model = load_model("./data/cnn_model.h5")
model.load_weights("./data/cnn_weights.h5")


@app.route("/")
def index():
    return render_template("index.html")


@app.route("/result", methods=['POST'])
def result():
    if request.files["image"]:
        img_file = request.files["image"]
        f = img_file.stream.read()
        bin_data = io.BytesIO(f)
        file_bytes = np.asarray(bytearray(bin_data.read()), dtype=np.uint8)
        img = cv2.imdecode(file_bytes, cv2.IMREAD_COLOR)
        img = cv2.resize(img, (32, 32))

        in_rows = 32
        in_cols = 32
        in_colors = 3

        labels = [
            'airplane',
            'automobile',
            'bird',
            'cat',
            'deer',
            'dog',
            'frog',
            'horse',
            'ship',
            'truck'
        ]

        img = img.reshape(-1, in_rows, in_cols,
                          in_colors).astype("float32") / 255

        r = model.predict(img, batch_size=32, verbose=1)
        res = r[0]
        return render_template("result.html", res=res)
    else:
        return redirect(url_for('index'))

index.html

<!DOCTYPE html>
<html>

<head>
    <title>アプリ</title>
    <link rel="stylesheet" type="text/css" href="../static/css/style.css">
</head>

<body class="main">

    <div class="heading">
        <h1>画像認識アプリ　Image-Recognition</h1>
    </div>


    <div>
        <form method="post" action="/result" enctype="multipart/form-data">
            <p>
                <input type="file" name="image" accept="image/png, image/jpeg" placeholder="Image Url">
            </p>
            <p>
                <input type="submit" value="Go!">
            </p>
        </form>
    </div>

</body>

</html>

result.html

<!DOCTYPE html>
<html>

<head>
    <title>Result</title>le.css
    <link rel="stylesheet" type="text/css" href="../static/css/style.css">
</head>

<body class="main">

    <div class="heading">
        予測結果は...<h1 align="center">{{labels[res.argmax()]}}</h1>
    </div>

</body>

</html>

行動規範の内容に同意します

回答1件

ベストアンサー

Flask and Keras model Error ''_thread._local' object has no attribute 'value''?に記載されているとおり、tensorflowのバージョンが原因だと思われます。

ざっと以下のような手順にて当方環境では動作することが確認できました。

前準備としてkeras/examples/cifar10_cnn.pyを使ってGoogle colab上で学習済みデータ(.h5)を作成。
ついでに指定した画像を予測するだけのコードをGoogle colab上で正常に動作するか確認。
python=3.6.9, tensoflow=1.15.0, keras=2.2.5の環境を構築。Flaskなどその他は最新。
raise ValueError("Tensor %s is not an element of this graph." % objが出たので

django+keras+apache+mod_wsgiで<tensor> is not an element of this graphが出た話。を参考に修正。

result.htmlにlevelsを渡すようにする。

参考まで。

予測データ作成

Python
1'''
2#Train a simple deep CNN on the CIFAR10 small images dataset.
3It gets to 75% validation accuracy in 25 epochs, and 79% after 50 epochs.
4(it's still underfitting at that point, though).
5'''
6import keras
7from keras.datasets import cifar10
8from keras.preprocessing.image import ImageDataGenerator
9from keras.models import Sequential
10from keras.layers import Dense, Dropout, Activation, Flatten
11from keras.layers import Conv2D, MaxPooling2D
12import os
13
14batch_size = 32
15num_classes = 10
16epochs = 100
17data_augmentation = True
18num_predictions = 20
19#save_dir = os.path.join(os.getcwd(), 'saved_models')
20save_dir = "."
21model_name = 'cnn_model.h5'
22w_name = 'cnn_weights.h5'
23
24# The data, split between train and test sets:
25(x_train, y_train), (x_test, y_test) = cifar10.load_data()
26print('x_train shape:', x_train.shape)
27print(x_train.shape[0], 'train samples')
28print(x_test.shape[0], 'test samples')
29
30# Convert class vectors to binary class matrices.
31y_train = keras.utils.to_categorical(y_train, num_classes)
32y_test = keras.utils.to_categorical(y_test, num_classes)
33
34model = Sequential()
35model.add(Conv2D(32, (3, 3), padding='same',
36                 input_shape=x_train.shape[1:]))
37model.add(Activation('relu'))
38model.add(Conv2D(32, (3, 3)))
39model.add(Activation('relu'))
40model.add(MaxPooling2D(pool_size=(2, 2)))
41model.add(Dropout(0.25))
42
43model.add(Conv2D(64, (3, 3), padding='same'))
44model.add(Activation('relu'))
45model.add(Conv2D(64, (3, 3)))
46model.add(Activation('relu'))
47model.add(MaxPooling2D(pool_size=(2, 2)))
48model.add(Dropout(0.25))
49
50model.add(Flatten())
51model.add(Dense(512))
52model.add(Activation('relu'))
53model.add(Dropout(0.5))
54model.add(Dense(num_classes))
55model.add(Activation('softmax'))
56
57# initiate RMSprop optimizer
58opt = keras.optimizers.RMSprop(lr=0.0001, decay=1e-6)
59
60# Let's train the model using RMSprop
61model.compile(loss='categorical_crossentropy',
62              optimizer=opt,
63              metrics=['accuracy'])
64
65x_train = x_train.astype('float32')
66x_test = x_test.astype('float32')
67x_train /= 255
68x_test /= 255
69
70if not data_augmentation:
71    print('Not using data augmentation.')
72    model.fit(x_train, y_train,
73              batch_size=batch_size,
74              epochs=epochs,
75              validation_data=(x_test, y_test),
76              shuffle=True)
77else:
78    print('Using real-time data augmentation.')
79    # This will do preprocessing and realtime data augmentation:
80    datagen = ImageDataGenerator(
81        featurewise_center=False,  # set input mean to 0 over the dataset
82        samplewise_center=False,  # set each sample mean to 0
83        featurewise_std_normalization=False,  # divide inputs by std of the dataset
84        samplewise_std_normalization=False,  # divide each input by its std
85        zca_whitening=False,  # apply ZCA whitening
86        zca_epsilon=1e-06,  # epsilon for ZCA whitening
87        rotation_range=0,  # randomly rotate images in the range (degrees, 0 to 180)
88        # randomly shift images horizontally (fraction of total width)
89        width_shift_range=0.1,
90        # randomly shift images vertically (fraction of total height)
91        height_shift_range=0.1,
92        shear_range=0.,  # set range for random shear
93        zoom_range=0.,  # set range for random zoom
94        channel_shift_range=0.,  # set range for random channel shifts
95        # set mode for filling points outside the input boundaries
96        fill_mode='nearest',
97        cval=0.,  # value used for fill_mode = "constant"
98        horizontal_flip=True,  # randomly flip images
99        vertical_flip=False,  # randomly flip images
100        # set rescaling factor (applied before any other transformation)
101        rescale=None,
102        # set function that will be applied on each input
103        preprocessing_function=None,
104        # image data format, either "channels_first" or "channels_last"
105        data_format=None,
106        # fraction of images reserved for validation (strictly between 0 and 1)
107        validation_split=0.0)
108
109    # Compute quantities required for feature-wise normalization
110    # (std, mean, and principal components if ZCA whitening is applied).
111    datagen.fit(x_train)
112
113    # Fit the model on the batches generated by datagen.flow().
114    model.fit_generator(datagen.flow(x_train, y_train,
115                                     batch_size=batch_size),
116                        epochs=epochs,
117                        validation_data=(x_test, y_test),
118                        workers=4)
119
120# Save model and weights
121if not os.path.isdir(save_dir):
122    os.makedirs(save_dir)
123model_path = os.path.join(save_dir, model_name)
124model.save(model_path)
125print('Saved trained model at %s ' % model_path)
126
127w_path = os.path.join(save_dir, w_name)
128model.save_weights(w_path)
129
130# Score trained model.
131scores = model.evaluate(x_test, y_test, verbose=1)
132print('Test loss:', scores[0])
133print('Test accuracy:', scores[1])

環境構築とテスト実行

>conda create -n cifar10 python=3.6.9 tensorflow=1.15.0 keras=2.2.5 matplotlib flask
>conda activate cifar10
(VS2015ビルド環境がないと怒られるが無視)
>conda install -c conda-forge opencv
>python app.py
(いろいろ警告が出るが無視)

app.py

Python
1import pickle
2import numpy as np
3from flask import Flask, render_template, request, redirect, url_for
4
5from keras.models import load_model
6import cv2
7import numpy as np
8import matplotlib.pyplot as plt
9import io
10import tensorflow as tf
11import sys
12
13app = Flask(__name__)
14
15# 機械学習モデルの読込
16model = load_model("./templates/cnn_model.h5")
17model.load_weights("./templates/cnn_weights.h5")
18
19graph = tf.get_default_graph()
20
21@app.route("/")
22def index():
23    return render_template("index.html")
24
25@app.route("/result", methods=['POST'])
26def result():
27    if request.files["image"]:
28        img_file = request.files["image"]
29        f = img_file.stream.read()
30        bin_data = io.BytesIO(f)
31        file_bytes = np.asarray(bytearray(bin_data.read()), dtype=np.uint8)
32
33        img = cv2.imdecode(file_bytes, cv2.IMREAD_COLOR)
34        img = cv2.resize(img, (32, 32))
35
36        in_rows = 32
37        in_cols = 32
38        in_colors = 3
39
40        labels = [
41            'airplane',
42            'automobile',
43            'bird',
44            'cat',
45            'deer',
46            'dog',
47            'frog',
48            'horse',
49            'ship',
50            'truck'
51        ]
52
53        img = img.reshape(-1, in_rows, in_cols,
54                          in_colors).astype("float32") / 255
55
56        with graph.as_default():
57            r = model.predict(img, batch_size=32, verbose=1)
58            res = r[0]
59            print(res, file=sys.stdout)
60        return render_template("result.html", res=res, labels=labels)
61    else:
62        return redirect(url_for('index'))
63
64if __name__ == '__main__':
65    app.run(debug=True)

投稿2020/02/16 01:37

編集2020/02/16 02:03

can110

総合スコア38278

masa.taka

2020/02/16 02:23 編集

非常にご親切な回答、誠にありがとうございます。ご助言を踏まえ、上記pyファイルを反映しましたところ、以下のエラーが発生しております。たびたび恐縮でございますが、解決策等、ございますでしょうか。 tensorflow.python.framework.errors_impl.FailedPreconditionError: Error while reading resource variable conv2d_10/kernel from Container: localhost. This could mean that the variable was uninitialized. Not found: Container localhost does not exist. (Could not find resource: localhost/conv2d_10/kernel) [[{{node conv2d_10/convolution/ReadVariableOp}}]]