VGG16 训练哪些事

1、出现caused by subtracting 2 from 1 for 'block1_pool/MaxPool

  File "D:\tool\learning_tools\python3.5\lib\site-packages\tensorflow\python\framework\common_shapes.py", line 627, in call_cpp_shape_fn
    require_shape_fn)
  File "D:\tool\learning_tools\python3.5\lib\site-packages\tensorflow\python\framework\common_shapes.py", line 691, in _call_cpp_shape_fn_impl
    raise ValueError(err.message)
ValueError: Negative dimension size caused by subtracting 2 from 1 for 'block1_pool/MaxPool' (op: 'MaxPool') with input shapes: [?,1,100,64].

错误的原因Keras是Theano或Tensorflow库的包装器在使用调用api出现的一些问题

1、~/.keras/keras.json设置image_dim_ordering
2、image_dim_ordering在模型中指定如下：model.add(MaxPooling2D(pool_size=(2, 2), dim_ordering="th"))
3、在程序中直接添加
    from keras import backend as K
    K.set_image_dim_ordering('th')

2、model.compile参数设置

# 二分类
#model.compile(loss='binary_crossentropy',
#              optimizer='rmsprop',
#              metrics=['accuracy'])

# 多分类
model.compile(loss='categorical_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])
# 优化器rmsprop：除学习率可调整外，建议保持优化器的其他默认参数不变

# 优化器的设置
# initiate RMSprop optimizer
opt = keras.optimizers.rmsprop(lr=0.0001, decay=1e-6)
model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.SGD(lr=1e-4, momentum=0.9),
              metrics=['accuracy'])

3、图片预处理

imageDataGenerator：
    用以生成一个batch的图像数据，支持实时数据提升。训练时该函数会无限生成数据，直到达到规定的epoch次数为止。
flow_from_directory(directory):
    以文件夹路径为参数,生成经过数据提升/归一化后的数据,在一个无限循环中无限产生batch数据，直接产生数据和标签。

from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale=1./255,
                                   shear_range=0.2,
                                   zoom_range=0.2,
                                   horizontal_flip=True)
'''
rescale: rescaling factor. Defaults to None.
	If None or 0, no rescaling is applied,
	otherwise we multiply the data by the value provided
	(after applying all other transformations).
shear_range: Float. Shear Intensity
	(Shear angle in counter-clockwise direction in degrees)
zoom_range: Float or [lower, upper]. Range for random zoom.
	If a float, `[lower, upper] = [1-zoom_range, 1+zoom_range]`
horizontal_flip: Boolean. Randomly flip inputs horizontally.
'''

flow_from_directory的用法介绍
    https://blog.csdn.net/mieleizhi0522/article/details/82191331

4、模型的fit的介绍

model.fit_generator(train_generator,
                    steps_per_epoch=1000,
                    epochs=2,
                    validation_data=validation_generator,
                    validation_steps=200)

model.fit(X_train,Y_train,epochs = 20,batch_size = 128)

 

 

 

 

 

转载于:https://my.oschina.net/u/3209854/blog/3054013