如何为图像集安装两个 keras ImageDataGenerator

Question

我正在为以下任务寻找解决方案或示例：

我有一组从不同角度拍摄的相同对象的图像。 我想用 keras 构建一个深度 CNN，它可以采用 sets 两张图像，分别对每张图像执行数据增强，并将它们一起输入到一个连接的模型中。

更详细的解释：

图像存储在 HDF5 文件中，具有以下形状：

data['Xp'] # shape=(3000, 224, 224, 3) #RGB images
data['Xs'] # shape=(3000, 224, 224, 3) #RGB images
data['Y']  # shape=(3000, 9) #categorical data.

现在，我想要一个生成器，它可以：

1. 打乱数据集的索引。
2. 然后拍照并
3. 来自数据的类别增加来自X1_train、X2_train的图像
4. 单独将其输入到具有以下结构的网络中：

代码...

from keras.layers import Flatten, Dense, Input, Dropout, Convolution2D, MaxPooling2D, Merge


img_input = Input(shape=input_shape)
x = Convolution2D(64, 3, 3, activation='relu', border_mode='same', name='block1_conv1')(img_input)
x = Convolution2D(64, 3, 3, activation='relu', border_mode='same', name='block1_conv2')(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)

# ... more network definition here ....

model1 = Model(img_input, x)
model2 = Model(img_input, x)

merged = Merge([model1, model2], mode='concat')

final_model = Sequential()
final_model.add(merged)
final_model.add(Dense(9, activation='softmax'))

我创建了以下生成器，它可以生成预期的数据来提供给 fit_generator的模特...

def aug_train_iterator(Xp, Xs, Y, database_file=database_file, is_binary=True):
    from itertools import izip
    from keras.preprocessing.image import ImageDataGenerator

    seed = 7 #make sure that two iterators give same tomato each time...

    ig = ImageDataGenerator(dim_ordering='tf', rotation_range=90,
                                               width_shift_range=0.05,
                                               height_shift_range=0.05,
                                               zoom_range=0.05,
                                               fill_mode='constant',
                                               cval=0.0,
                                               horizontal_flip=True,
                                               rescale=1./255)


    for batch in izip(ig.flow(Xp,Y,  seed=seed), ig.flow(Xs, seed=seed)):
        for i in range(len(batch[0][0])):
            x1 = batch[0][0][i].reshape(1,224, 224, 3)
            x2 = batch[1][i].reshape(1, 224, 224, 3)
            y = batch[0][1][i].reshape(1,2)
            yield ([x1, x2], y)

现在，当我尝试拟合模型时...

gen = aug_train_iterator(Xp, Xs, Y)
final_model.fit_generator(gen, 1000, 20)

它实际上运行了几张图片......然后引发了大约 15 张图片的错误：

Epoch 1/20
  15/1000 [..............................] - ETA: 606s - loss: 0.7001 - acc: 0.4000

Exception in thread Thread-44:
Traceback (most recent call last):
  File "/usr/lib/python2.7/threading.py", line 810, in __bootstrap_inner
    self.run()
  File "/usr/lib/python2.7/threading.py", line 763, in run
    self.__target(*self.__args, **self.__kwargs)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 404, in data_generator_task
    generator_output = next(generator)
  File "<ipython-input-134-f128a127c7ce>", line 35, in aug_train_iterator
    for batch in izip(ig.flow(Xp,Y,  seed=seed), ig.flow(Xs, seed=seed)):
  File "/usr/local/lib/python2.7/dist-packages/keras/preprocessing/image.py", line 495, in next
    x = self.X[j]
  File "/usr/lib/python2.7/dist-packages/h5py/_hl/dataset.py", line 367, in __getitem__
    if self._local.astype is not None:
AttributeError: 'thread._local' object has no attribute 'astype'

有什么问题？

Answer 1

使用 pickle_safe=True 解决了问题