基于CIFAR10(小批量图片)数据集训练简单的胶囊(组神经元)深度卷积神经网络
代码注释
"""Train a simple CNN-Capsule Network on the CIFAR10 small images dataset.
基于CIFAR10(小批量图片)数据集训练简单的胶囊(组神经元)深度卷积神经网络
Without Data Augmentation:
It gets to 75% validation accuracy in 10 epochs,
and 79% after 15 epochs, and overfitting after 20 epochs.
不扩展数据集,10个周期后达到7%%的准确率,15个周期后达到79%的准确率,20个周期后过拟合。
With Data Augmentation:
It gets to 75% validation accuracy in 10 epochs,
and 79% after 15 epochs, and 83% after 30 epcohs.
In my test, highest validation accuracy is 83.79% after 50 epcohs.
不扩展数据集,10个周期后达到7%%的准确率,15个周期后达到79%的准确率,30个周期达到83%的准确率。
测试中,最好成绩是50个周期达到83.79%的准确率
This is a fast Implement, just 20s/epcoh with a gtx 1070 gpu.
快速实现,基于一个gtx 1070 GPU(图像处理器,显卡)每个周期执行需要20秒
"""
from __future__ import print_function
from keras import backend as K
from keras.engine.topology import Layer
from keras import activations
from keras import utils
from keras.datasets import cifar10
from keras.models import Model
from keras.layers import *
from keras.preprocessing.image import ImageDataGenerator
# the squashing function.
# 压平函数
# we use 0.5 in stead of 1 in hinton's paper.
# 在hinton的论文中,使用0.5代替1
# if 1, the norm of vector will be zoomed out.
# 如果是1,向量范数将缩小
# if 0.5, the norm will be zoomed in while original norm is less than 0.5
# 如果是1,原范数小于0.5,向量范数将放大
# and be zoomed out while original norm is greater than 0.5.
# 原范数大于0.5,向量范数将缩小
def squash(x, axis=-1):
s_squared_norm = K.sum(K.square(x), axis, keepdims=True) + K.epsilon()
scale = K.sqrt(s_squared_norm) / (0.5 + s_squared_norm)
return scale * x
# define our own softmax function instead of K.softmax
# because K.softmax can not specify axis.
# 自定义softmax函数,替换K.softmax函数,因为K.softmax函数不能指定轴
def softmax(x, axis=-1):
ex = K.exp(x - K.max(x, axis=axis, keepdims=True))
return ex / K.sum(ex, axis=axis, keepdims=True)
# define the margin loss like hinge loss
# 定义利润边缘损失,如hinge损失
def margin_loss(y_true, y_pred):
lamb, margin = 0.5, 0.1
return y_true * K.square(K.relu(1 - margin - y_pred)) + lamb * (
1 - y_true) * K.square(K.relu(y_pred - margin))
class Capsule(Layer):
"""A Capsule Implement with Pure Keras
基于纯Keras的胶囊(组神经元)实现
There are two vesions of Capsule.
以下是2个版本的胶囊(组神经元)
One is like dense layer (for the fixed-shape input),
一种类似与全连接层(对于固定形状输入),
and the other is like timedistributed dense (for various length input).
一种类似时间分布的全连接层(对于变长输入),
The input shape of Capsule must be (batch_size,
input_num_capsule,
input_dim_capsule
)
胶囊(组神经元)输入(数据)形状为:
and the output shape is (batch_size,
num_capsule,
dim_capsule
)
输出(数据)形状为:
Capsule Implement is from https://github.com/bojone/Capsule/
胶囊(组神经元)实现见:https://github.com/bojone/Capsule/
Capsule Paper: https://arxiv.org/abs/1710.09829
胶囊(组神经元)论文:https://arxiv.org/abs/1710.09829
"""
def __init__(self,
num_capsule,
dim_capsule,
routings=3,
share_weights=True,
activation='squash',
**kwargs):
super(Capsule, self).__init__(**kwargs)
self.num_capsule = num_capsule
self.dim_capsule = dim_capsule
self.routings = routings
self.share_weights = share_weights
if activation == 'squash':
self.activation = squash
else:
self.activation = activations.get(activation)
def build(self, input_shape):
input_dim_capsule = input_shape[-1]
if self.share_weights:
self.kernel = self.add_weight(
name='capsule_kernel',
shape=(1, input_dim_capsule,
self.num_capsule * self.dim_capsule),
initializer='glorot_uniform',
trainable=True)
else:
input_num_capsule = input_shape[-2]
self.kernel = self.add_weight(
name='capsule_kernel',
shape=(input_num_capsule, input_dim_capsule,
self.num_capsule * self.dim_capsule),
initializer='glorot_uniform',
trainable=True)
def call(self, inputs):
"""Following the routing algorithm from Hinton's paper,
根据Hinton论文的路由算法,
but replace b = b + <u,v> with b = <u,v>.
但是 用b = <u,v>替换b = b + <u,v>
This change can improve the feature representation of Capsule.
这种改变可以改善胶囊(组神经元)的特征表示。
However, you can replace
而且,可以把替换
b = K.batch_dot(outputs, hat_inputs, [2, 3])
with
使用
b += K.batch_dot(outputs, hat_inputs, [2, 3])
to realize a standard routing.
实现标准路由。
"""
if self.share_weights:
hat_inputs = K.conv1d(inputs, self.kernel)
else:
hat_inputs = K.local_conv1d(inputs, self.kernel, [1], [1])
batch_size = K.shape(inputs)[0]
input_num_capsule = K.shape(inputs)[1]
hat_inputs = K.reshape(hat_inputs,
(batch_size, input_num_capsule,
self.num_capsule, self.dim_capsule))
hat_inputs = K.permute_dimensions(hat_inputs, (0, 2, 1, 3))
b = K.zeros_like(hat_inputs[:, :, :, 0])
for i in range(self.routings):
c = softmax(b, 1)
if K.backend() == 'theano':
o = K.sum(o, axis=1)
o = self.activation(K.batch_dot(c, hat_inputs, [2, 2]))
if i < self.routings - 1:
b = K.batch_dot(o, hat_inputs, [2, 3])
if K.backend() == 'theano':
o = K.sum(o, axis=1)
return o
def compute_output_shape(self, input_shape):
return (None, self.num_capsule, self.dim_capsule)
batch_size = 128
num_classes = 10
epochs = 100
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
y_train = utils.to_categorical(y_train, num_classes)
y_test = utils.to_categorical(y_test, num_classes)
# A common Conv2D model
# 2维卷积模型
input_image = Input(shape=(None, None, 3))
x = Conv2D(64, (3, 3), activation='relu')(input_image)
x = Conv2D(64, (3, 3), activation='relu')(x)
x = AveragePooling2D((2, 2))(x)
x = Conv2D(128, (3, 3), activation='relu')(x)
x = Conv2D(128, (3, 3), activation='relu')(x)
"""now we reshape it as (batch_size, input_num_capsule, input_dim_capsule)
then connect a Capsule layer.
重塑(数据)形状为 (batch_size, input_num_capsule, input_dim_capsule)
然后,链接一个胶囊(组神经元)层
the output of final model is the lengths of 10 Capsule, whose dim=16.
最终模型输出为16维的10个胶囊(组神经元)
the length of Capsule is the proba,
胶囊(组神经元)的长度是proba
so the problem becomes a 10 two-classification problem.
因此问题成为10个两分类问题。
"""
x = Reshape((-1, 128))(x)
capsule = Capsule(10, 16, 3, True)(x)
output = Lambda(lambda z: K.sqrt(K.sum(K.square(z), 2)))(capsule)
model = Model(inputs=input_image, outputs=output)
# we use a margin loss
# 使用边缘损失(函数)
model.compile(loss=margin_loss, optimizer='adam', metrics=['accuracy'])
model.summary()
# we can compare the performance with or without data augmentation
# 比较效果,有和没有数据集(扩大)
data_augmentation = True
if not data_augmentation:
print('Not using data augmentation.')
model.fit(
x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
validation_data=(x_test, y_test),
shuffle=True)
else:
print('Using real-time data augmentation.')
# This will do preprocessing and realtime data augmentation:
# 预处理和实时数据扩大(通过平移、翻转等图像变换增加图像样本数量)。
datagen = ImageDataGenerator(
featurewise_center=False, # set input mean to 0 over the dataset # 基于数据集,使输入数据平均值为0
samplewise_center=False, # set each sample mean to 0 # 使样本平均值为0
featurewise_std_normalization=False, # divide inputs by std of the dataset # 通过数据标准化划分输入数据
samplewise_std_normalization=False, # divide each input by its std # 通过标准化划分输入数据
zca_whitening=False, # apply ZCA(Zero-phase Component Analysis) whitening # 对输入数据施加ZCA白化
rotation_range=0, # randomly rotate images in the range (degrees, 0 to 180) # 旋转图像0-180度
width_shift_range=0.1, # randomly shift images horizontally (fraction of total width) # 水平平移图像(基于图像宽度比例)
height_shift_range=0.1, # randomly shift images vertically (fraction of total height) # 垂直平移图像(基于图像高度比例)
horizontal_flip=True, # randomly flip images # 水平翻转图像
vertical_flip=False) # randomly flip images # 垂直翻转图像
# Compute quantities required for feature-wise normalization
# 特征归一化的计算量
# (std, mean, and principal components if ZCA whitening is applied).
# (如果ZCA白化(一种降维方法)会使用标准化、均值和主成分方法)
datagen.fit(x_train)
# Fit the model on the batches generated by datagen.flow().
# 使用datagen.flow()生成的批次数据在模型训练
model.fit_generator(
datagen.flow(x_train, y_train, batch_size=batch_size),
epochs=epochs,
validation_data=(x_test, y_test),
workers=4)
代码执行
Keras详细介绍
中文:http://keras-cn.readthedocs.io/en/latest/
实例下载
https://github.com/keras-team/keras
https://github.com/keras-team/keras/tree/master/examples
完整项目下载
方便没积分童鞋,请加企鹅452205574,共享文件夹。
包括:代码、数据集合(图片)、已生成model、安装库文件等。