图像识别笔记

图像识别笔记

CNN

‘
’
‘

VGG16

‘
’
‘

ResNet

• 包含两篇论文Deep Residual learning for image recognition 和 training very deep network

‘
’
‘

Vgg16代码

• 数据是keras自带的CIFAR100小图像分类集，参考这里
• 相关代码参考Vgg16和Resnet
• vgg16每次卷积前要padding，filter是3*3，所以ZeroPadding2D(padding=(1,1)),这个操作和在Conv2D中设置padding='same’一样
• 每次卷积后要在channel的维度做 BatchNormalization

# -*- coding: utf-8 -*-
"""
Created on Wed Aug 19 15:18:58 2020

@author: 86188
"""
'''
x_train:(50000, 32, 32, 3)

y_train.shape:(50000, 1)
'''
import tensorflow as tf
from keras.datasets import cifar10
from keras.models import Model
from keras.layers import Masking, ZeroPadding2D, Bidirectional,Dense, Input, Conv2D, Activation,Dropout,MaxPooling2D,BatchNormalization,Flatten
#from keras.preprocessing import sequence
from keras import metrics
from keras import backend as K
from keras.callbacks import Callback
#import matplotlib.pyplot as plt
from keras.callbacks import ReduceLROnPlateau
from keras.callbacks import EarlyStopping
from keras.utils.np_utils import to_categorical


‘’‘
数据是keras自带的
‘’‘
(x_train, y_train), (x_test, y_test) = cifar10.load_data() #

y_train = to_categorical(y_train, num_classes=10)
y_test = to_categorical(y_test, num_classes=10)

'''
Conv2D(nb_filter, kernel_size, padding=padding, strides=strides, activation='relu', name=conv_name,padding='same')(x)
data_forma="channels_last"(默认)
padding='same':默认pad到不改变图大小,padding='valid'没有padding
channels_last 对应输入尺寸为 (batch, height, width, channels)，
channels_first 对应输入尺寸为 (batch, channels, height, width)
'''
def Conv2d_BN(x, nb_filter, kernel_size, strides=(1, 1), padding='same', name=None):
#    if name is not None:
#        bn_name = name + '_bn'
#        conv_name = name + '_conv'
#    else:
#        bn_name = None
#        conv_name = None
 
    x = Conv2D(nb_filter, kernel_size, padding=padding, strides=strides, activation='relu')(x)
    x = BatchNormalization(axis=3)(x)
    return x


def VGG_16(weights_path=None):


    pic_input = Input(shape=(32,32,3))
    x = ZeroPadding2D((1, 1))(pic_input)
    
    x=Conv2d_BN(x, nb_filter=64, kernel_size=(3,3), padding='valid')
    x=Conv2d_BN(x, nb_filter=64, kernel_size=(3,3), padding='same')
    x=MaxPooling2D((2,2),strides=(2,2))(x)


    x=Conv2d_BN(x, nb_filter=128, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=128, kernel_size=(3,3), padding='same')
    x=MaxPooling2D((2,2),strides=(2,2))(x)
    
    x=Conv2d_BN(x, nb_filter=256, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=256, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=256, kernel_size=(3,3), padding='same')
    x=MaxPooling2D((2,2),strides=(2,2))(x)
 
    x=Conv2d_BN(x, nb_filter=512, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=512, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=512, kernel_size=(3,3), padding='same')
    x=MaxPooling2D((2,2),strides=(2,2))(x)
 
    x=Conv2d_BN(x, nb_filter=512, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=512, kernel_size=(3,3), padding='same')
    x=Conv2d_BN(x, nb_filter=512, kernel_size=(3,3), padding='same')
    x=MaxPooling2D((2,2),strides=(2,2))(x)
    
    x=Flatten()(x)
    x=Dense(4096, activation='relu')(x)
    x=Dropout(0.5)(x)
    x=Dense(4096, activation='relu')(x)
    x=Dropout(0.5)(x) 
    x=Dense(10, activation='softmax')(x)

    model = Model(inputs=pic_input, outputs=x)
    model.summary()
 
    if weights_path:
        model.load_weights(weights_path)
 
    return model

reduce_lr = ReduceLROnPlateau(monitor='val_loss', patience=3,factor=0.1,mode='min')    
early_stopping = EarlyStopping(monitor='val_loss', patience=5, verbose=2, mode='min')
model=VGG_16()
model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['accuracy'])
hist1=model.fit(x_train, y_train, validation_split=0.1,epochs=10, batch_size=32,callbacks=[reduce_lr,early_stopping])



score=model.evaluate(x_test,y_test,batch_size=32,verbose=2)
model.save("vgg_try.h5")

#def VGG_16(weights_path=None):
#    model = Sequential()
#    model.add(ZeroPadding2D((1,1),input_shape=(32,32,3)))
#    model.add(Convolution2D(64, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(64, 3, 3, activation='relu'))
#    model.add(MaxPooling2D((2,2), strides=(2,2)))
# 
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(128, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(128, 3, 3, activation='relu'))
#    model.add(MaxPooling2D((2,2), strides=(2,2)))
# 
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(256, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(256, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(256, 3, 3, activation='relu'))
#    model.add(MaxPooling2D((2,2), strides=(2,2)))
# 
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(512, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(512, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(512, 3, 3, activation='relu'))
#    model.add(MaxPooling2D((2,2), strides=(2,2)))
# 
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(512, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(512, 3, 3, activation='relu'))
#    model.add(ZeroPadding2D((1,1)))
#    model.add(Convolution2D(512, 3, 3, activation='relu'))
#    model.add(MaxPooling2D((2,2), strides=(2,2)))
# 
#    model.add(Flatten())
#    model.add(Dense(4096, activation='relu'))
#    model.add(Dropout(0.5))
#    model.add(Dense(4096, activation='relu'))
#    model.add(Dropout(0.5))
#    model.add(Dense(1000, activation='softmax'))
# 
#    if weights_path:
#        model.load_weights(weights_path)
# 
#    return model