keras和tensorflow模型同時讀取報錯如何解決

這篇文章主要介紹“keras和tensorflow模型同時讀取報錯如何解決”的相關知識，小編通過實際案例向大家展示操作過程，操作方法簡單快捷，實用性強，希望這篇“keras和tensorflow模型同時讀取報錯如何解決”文章能幫助大家解決問題。

在使用tensorflow與keras混用是model.save 是正常的但是在load_model的時候報錯了在這里mark 一下

其中錯誤為：TypeError: tuple indices must be integers, not list

再一一番百度后無結果，上谷歌后找到了類似的問題。但是是一對鳥文不知道什么東西（翻譯后發現是俄文）。后來谷歌翻譯了一下找到了解決方法。故將原始問題文章貼上來警示一下

原訓練代碼

from tensorflow.python.keras.preprocessing.image import ImageDataGenerator
from tensorflow.python.keras.models import Sequential
from tensorflow.python.keras.layers import Conv2D, MaxPooling2D, BatchNormalization
from tensorflow.python.keras.layers import Activation, Dropout, Flatten, Dense
 
#Каталог с данными для обучения
train_dir = 'train'
# Каталог с данными для проверки
val_dir = 'val'
# Каталог с данными для тестирования
test_dir = 'val'
 
# Размеры изображения
img_width, img_height = 800, 800
# Размерность тензора на основе изображения для входных данных в нейронную сеть
# backend Tensorflow, channels_last
input_shape = (img_width, img_height, 3)
# Количество эпох
epochs = 1
# Размер мини-выборки
batch_size = 4
# Количество изображений для обучения
nb_train_samples = 300
# Количество изображений для проверки
nb_validation_samples = 25
# Количество изображений для тестирования
nb_test_samples = 25
 
model = Sequential()
 
model.add(Conv2D(32, (7, 7), padding="same", input_shape=input_shape))
model.add(BatchNormalization())
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(10, 10)))
 
model.add(Conv2D(64, (5, 5), padding="same"))
model.add(BatchNormalization())
model.add(Activation('tanh'))
model.add(MaxPooling2D(pool_size=(10, 10)))
 
model.add(Flatten())
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))
 
model.compile(loss='categorical_crossentropy',
              optimizer="Nadam",
              metrics=['accuracy'])
print(model.summary())
datagen = ImageDataGenerator(rescale=1. / 255)
 
train_generator = datagen.flow_from_directory(
    train_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='categorical')
 
val_generator = datagen.flow_from_directory(
    val_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='categorical')
 
test_generator = datagen.flow_from_directory(
    test_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='categorical')
 
model.fit_generator(
    train_generator,
    steps_per_epoch=nb_train_samples // batch_size,
    epochs=epochs,
    validation_data=val_generator,
    validation_steps=nb_validation_samples // batch_size)
 
print('Сохраняем сеть')
 
model.save("grib.h6")
print("Сохранение завершено!")

模型載入

from tensorflow.python.keras.preprocessing.image import ImageDataGenerator
from tensorflow.python.keras.models import Sequential
from tensorflow.python.keras.layers import Conv2D, MaxPooling2D, BatchNormalization
from tensorflow.python.keras.layers import Activation, Dropout, Flatten, Dense
from keras.models import load_model
 
print("Загрузка сети")
model = load_model("grib.h6")
print("Загрузка завершена!")

報錯

/usr/bin/python3.5 /home/disk2/py/neroset/do.py
/home/mama/.local/lib/python3.5/site-packages/h6py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
 from ._conv import register_converters as _register_converters
Using TensorFlow backend.
Загрузка сети
Traceback (most recent call last):
 File "/home/disk2/py/neroset/do.py", line 13, in <module>
model = load_model("grib.h6")
 File "/usr/local/lib/python3.5/dist-packages/keras/models.py", line 243, in load_model
model = model_from_config(model_config, custom_objects=custom_objects)
 File "/usr/local/lib/python3.5/dist-packages/keras/models.py", line 317, in model_from_config
return layer_module.deserialize(config, custom_objects=custom_objects)
 File "/usr/local/lib/python3.5/dist-packages/keras/layers/__init__.py", line 55, in deserialize
printable_module_name='layer')
 File "/usr/local/lib/python3.5/dist-packages/keras/utils/generic_utils.py", line 144, in deserialize_keras_object
list(custom_objects.items())))
 File "/usr/local/lib/python3.5/dist-packages/keras/models.py", line 1350, in from_config
model.add(layer)
 File "/usr/local/lib/python3.5/dist-packages/keras/models.py", line 492, in add
output_tensor = layer(self.outputs[0])
 File "/usr/local/lib/python3.5/dist-packages/keras/engine/topology.py", line 590, in __call__
self.build(input_shapes[0])
 File "/usr/local/lib/python3.5/dist-packages/keras/layers/normalization.py", line 92, in build
dim = input_shape[self.axis]
TypeError: tuple indices must be integers or slices, not list

Process finished with exit code 1

戰斗種族解釋

убераю BatchNormalization всё работает хорошо. Не подскажите в чём ошибка?Выяснил что сохранение keras и нормализация tensorflow не работают вместе нужно просто изменить строку импорта.（譯文：整理BatchNormalization一切正常。 不要告訴我錯誤是什么？我發現保存keras和規范化tensorflow不能一起工作;只需更改導入字符串即可。)

強調文本 強調文本

keras.preprocessing.image import ImageDataGenerator
keras.models import Sequential
keras.layers import Conv2D, MaxPooling2D, BatchNormalization
keras.layers import Activation, Dropout, Flatten, Dense

##完美解決

補充：keras和tensorflow模型同時讀取要慎重

項目中，先讀取了一個keras模型獲取模型輸入size，再加載keras轉tensorflow后的pb模型進行預測。

報錯：

Attempting to use uninitialized value batch_normalization_14/moving_mean

逛論壇，有建議加上初始化：

sess.run(tf.global_variables_initializer())

但是這樣的話，會導致模型參數全部變成初始化數據。無法使用預測模型參數。

最后發現，將keras模型的加載去掉即可。

猜測原因：keras模型和tensorflow模型同時讀取有坑

import cv2
import numpy as np
from keras.models import load_model
from utils.datasets import get_labels
from utils.preprocessor import preprocess_input
import time
import os
import tensorflow as tf
from tensorflow.python.platform import gfile
 
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
 
emotion_labels = get_labels('fer2013')
emotion_target_size = (64,64)
#emotion_model_path = './models/emotion_model.hdf5'
#emotion_classifier = load_model(emotion_model_path)
#emotion_target_size = emotion_classifier.input_shape[1:3]
 
path = '/mnt/nas/cv_data/emotion/test'
filelist = os.listdir(path)
total_num = len(filelist)
timeall = 0
n = 0
 
sess = tf.Session()
#sess.run(tf.global_variables_initializer())
with gfile.FastGFile("./trans_model/emotion_mode.pb", 'rb') as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())
    sess.graph.as_default()
    tf.import_graph_def(graph_def, name='')
 
    pred = sess.graph.get_tensor_by_name("predictions/Softmax:0")
 
    ######################img##########################
    for item in filelist:
        if (item == '.DS_Store') | (item == 'Thumbs.db'):
            continue
        src = os.path.join(os.path.abspath(path), item)
        bgr_image = cv2.imread(src)
        gray_image = cv2.cvtColor(bgr_image, cv2.COLOR_BGR2GRAY)
        gray_face = gray_image
        try:
            gray_face = cv2.resize(gray_face, (emotion_target_size))
        except:
            continue
 
        gray_face = preprocess_input(gray_face, True)
        gray_face = np.expand_dims(gray_face, 0)
        gray_face = np.expand_dims(gray_face, -1)
 
        input = sess.graph.get_tensor_by_name('input_1:0')
        res = sess.run(pred, {input: gray_face})
        print("src:", src)
 
        emotion_probability = np.max(res[0])
        emotion_label_arg = np.argmax(res[0])
        emotion_text = emotion_labels[emotion_label_arg]
        print("predict:", res[0], ",prob:", emotion_probability, ",label:", emotion_label_arg, ",text:",emotion_text)

關于“keras和tensorflow模型同時讀取報錯如何解決”的內容就介紹到這里了，感謝大家的閱讀。如果想了解更多行業相關的知識，可以關注億速云行業資訊頻道，小編每天都會為大家更新不同的知識點。