解决tensorflow 与keras 混用之坑
作者：Elijah_Yi
在使用tensorflow与keras混用是model.save 是正常的但是在load_model的时候报错了在这里mark 一下
其中错误为：TypeError: tuple indices must be integers, not list
再一一番百度后无结果，上谷歌后找到了类似的问题。但是是一对鸟文不知道什么东西（翻译后发现是俄文）。后来谷歌翻译了一下找到了解决方法。故将原始问题文章贴上来警示一下
原训练代码
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.h5")
print("Сохранение завершено!")
模型载入
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.h5")
print("Загрузка завершена!")
报错
/usr/bin/python3.5 /home/disk2/py/neroset/do.py
/home/mama/.local/lib/python3.5/site-packages/h5py/__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.h5")
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
战斗种族解释
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.（译文：整理BatchNormalization一切正常。 不要告诉我错误是什么？我发现保存keras和规范化tensorflow不能一起工作;只需更改导入字符串即可。)
强调文本 强调文本
keras.preprocessing.image import ImageDataGenerator keras.models import Sequential keras.layers import Conv2D, MaxPooling2D, BatchNormalization keras.layers import Activation, Dropout, Flatten, Dense
##完美解决
##附上原文链接
https://qa-help.ru/questions/keras-batchnormalization
补充：keras和tensorflow模型同时读取要慎重
项目中，先读取了一个keras模型获取模型输入size，再加载keras转tensorflow后的pb模型进行预测。
报错：
Attempting to use uninitialized value batch_normalization_14/moving_mean
逛论坛，有建议加上初始化：
sess.run(tf.global_variables_initializer())
但是这样的话，会导致模型参数全部变成初始化数据。无法使用预测模型参数。
最后发现，将keras模型的加载去掉即可。
猜测原因：keras模型和tensorflow模型同时读取有坑
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)
以上为个人经验，希望能给大家一个参考，也希望大家多多支持脚本之家。