Day04- interpretation of the classic convolution neural network

Day04- interpretation of the classic convolution neural network

Job Description

Today's real items is based on the "mask category" classic convolution neural network of VGG.

Masks recognition, is one that can effectively detect and carry the owner did not wear a mask carrying a dense crowd in the face area, while the judge whether to wear a mask. Usually consists of two functional units, the classification can be done to detect face masks and face masks, respectively.

The practice of production compared to the problems identified masks environment, reduce the difficulty to achieve only face masks judgment model, enabling determination of whether to wear face masks . This practice is designed to identify the text column through a mask, so that we understand and know how to use dynamic diagram to build a flying propeller classic convolution neural network.

Special Note: This practice data sets from the Internet, not for business purposes with.

Work requirements:

• 1, according to the contents learned in class, and run on a network constructed VGGNet. On this basis, try another network configuration.
• 2, thinking and hands-parameter adjustment, optimization, improve the accuracy of the test set.

Courseware and link data set is available to preface looking for introduction before the formal learning

All we need is the data contained in the day04 folder. characterData.zip is a data set that we need to use, CarID.png is used to test the effect of the final picture.

Sample Code

First, the environment configuration

# 导入需要的包

import os
import zipfile
import random
import json
import paddle
import sys
import numpy as np
from PIL import Image
from PIL import ImageEnhance
import paddle.fluid as fluid
from multiprocessing import cpu_count
import matplotlib.pyplot as plt

# 参数配置

train_parameters = {
    "input_size": [3, 224, 224],                              #输入图片的shape
    "class_dim": -1,                                          #分类数
    "src_path":"/home/aistudio/work/maskDetect.zip",#原始数据集路径
    "target_path":"/home/aistudio/data/",                     #要解压的路径
    "train_list_path": "/home/aistudio/data/train.txt",       #train.txt路径
    "eval_list_path": "/home/aistudio/data/eval.txt",         #eval.txt路径
    "readme_path": "/home/aistudio/data/readme.json",         #readme.json路径
    "label_dict":{},                                          #标签字典
    "num_epochs": 1,                                         #训练轮数
    "train_batch_size": 8,                                    #训练时每个批次的大小
    "learning_strategy": {                                    #优化函数相关的配置
        "lr": 0.001                                           #超参数学习率
    } 
}

Second, the data preparation

1. Decompressing the original data set
2. Divided in proportion to the training set and validation set
3. Scrambled, generating a data list
4. Configured to provide training data set and a validation data set provider

def unzip_data(src_path,target_path):
    '''
    解压原始数据集，将src_path路径下的zip包解压至data目录下
    '''
    if(not os.path.isdir(target_path + "maskDetect")):     
        z = zipfile.ZipFile(src_path, 'r')
        z.extractall(path=target_path)
        z.close()

def get_data_list(target_path,train_list_path,eval_list_path):
    '''
    生成数据列表
    '''
    #存放所有类别的信息
    class_detail = []
    #获取所有类别保存的文件夹名称
    data_list_path=target_path+"maskDetect/"
    class_dirs = os.listdir(data_list_path)  
    #总的图像数量
    all_class_images = 0
    #存放类别标签
    class_label=0
    #存放类别数目
    class_dim = 0
    #存储要写进eval.txt和train.txt中的内容
    trainer_list=[]
    eval_list=[]
    #读取每个类别，['maskimages', 'nomaskimages']
    for class_dir in class_dirs:
        if class_dir != ".DS_Store":
            class_dim += 1
            #每个类别的信息
            class_detail_list = {}
            eval_sum = 0
            trainer_sum = 0
            #统计每个类别有多少张图片
            class_sum = 0
            #获取类别路径 
            path = data_list_path  + class_dir
            # 获取所有图片
            img_paths = os.listdir(path)
            for img_path in img_paths:                                  # 遍历文件夹下的每个图片
                name_path = path + '/' + img_path                       # 每张图片的路径
                if class_sum % 10 == 0:                                 # 每10张图片取一个做验证数据
                    eval_sum += 1                                       # test_sum为测试数据的数目
                    eval_list.append(name_path + "\t%d" % class_label + "\n")
                else:
                    trainer_sum += 1 
                    trainer_list.append(name_path + "\t%d" % class_label + "\n")#trainer_sum测试数据的数目
                class_sum += 1                                          #每类图片的数目
                all_class_images += 1                                   #所有类图片的数目
             
            # 说明的json文件的class_detail数据
            class_detail_list['class_name'] = class_dir             #类别名称，如jiangwen
            class_detail_list['class_label'] = class_label          #类别标签
            class_detail_list['class_eval_images'] = eval_sum       #该类数据的测试集数目
            class_detail_list['class_trainer_images'] = trainer_sum #该类数据的训练集数目
            class_detail.append(class_detail_list)  
            #初始化标签列表
            train_parameters['label_dict'][str(class_label)] = class_dir
            class_label += 1 
            
    #初始化分类数
    train_parameters['class_dim'] = class_dim

   
    
    #乱序  
    random.shuffle(eval_list)
    with open(eval_list_path, 'a') as f:
        for eval_image in eval_list:
            f.write(eval_image) 
            
    random.shuffle(trainer_list)
    with open(train_list_path, 'a') as f2:
        for train_image in trainer_list:
            f2.write(train_image) 

    # 说明的json文件信息
    readjson = {}
    readjson['all_class_name'] = data_list_path                  #文件父目录
    readjson['all_class_images'] = all_class_images
    readjson['class_detail'] = class_detail
    jsons = json.dumps(readjson, sort_keys=True, indent=4, separators=(',', ': '))
    with open(train_parameters['readme_path'],'w') as f:
        f.write(jsons)
    print ('生成数据列表完成！')

def custom_reader(file_list):
    '''
    自定义reader
    '''
    def reader():
        with open(file_list, 'r') as f:
            lines = [line.strip() for line in f]
            for line in lines:
                img_path, lab = line.strip().split('\t')
                img = Image.open(img_path) 
                if img.mode != 'RGB': 
                    img = img.convert('RGB') 
                img = img.resize((224, 224), Image.BILINEAR)
                img = np.array(img).astype('float32') 
                img = img.transpose((2, 0, 1))  # HWC to CHW 
                img = img/255                # 像素值归一化 
                yield img, int(lab) 
    return reader

# 参数初始化

src_path=train_parameters['src_path']
target_path=train_parameters['target_path']
train_list_path=train_parameters['train_list_path']
eval_list_path=train_parameters['eval_list_path']
batch_size=train_parameters['train_batch_size']

'''
解压原始数据到指定路径
'''
unzip_data(src_path,target_path)

'''
划分训练集与验证集，乱序，生成数据列表
'''#每次生成数据列表前，首先清空train.txt和eval.txtwith open(train_list_path, 'w') as f: 
    f.seek(0)
    f.truncate() 
with open(eval_list_path, 'w') as f: 
    f.seek(0)
    f.truncate() 
#生成数据列表   
get_data_list(target_path,train_list_path,eval_list_path)

'''
构造数据提供器
'''
train_reader = paddle.batch(custom_reader(train_list_path),
                            batch_size=batch_size,
                            drop_last=True)
eval_reader = paddle.batch(custom_reader(eval_list_path),
                            batch_size=batch_size,
                            drop_last=True)

Third, the model configuration

VGG core group of five convolution operation between each of the two groups do Max-Pooling space dimension reduction. The number of times using the same group of successive 3X3 convolution, the convolution kernel increase of 64 to 512 shallow group, the number of convolution kernels in the same group is the same as the deepest group. After two full convolution contact connection layer, followed by a classification layer. Since the different layers within each convolution with these types of layers 11,13,16,19 model, on FIG. 16 shows a network structure layer.

class ConvPool(fluid.dygraph.Layer):
    '''卷积+池化'''
    def __init__(self,
                 num_channels,
                 num_filters,
                 filter_size,
                 pool_size,
                 pool_stride,
                 groups,
                 pool_padding=1,
                 pool_type='max',
                 conv_stride=1,
                 conv_padding=0,
                 act=None):
        super(ConvPool, self).__init__()  

        self._conv2d_list = []

        for i in range(groups):
            conv2d = self.add_sublayer(   #返回一个由所有子层组成的列表。
                'bb_%d' % i,
                fluid.dygraph.Conv2D(
                num_channels=num_channels, #通道数
                num_filters=num_filters,   #卷积核个数
                filter_size=filter_size,   #卷积核大小
                stride=conv_stride,        #步长
                padding=conv_padding,      #padding大小，默认为0
                act=act)
            )
        self._conv2d_list.append(conv2d)   

        self._pool2d = fluid.dygraph.Pool2D(
            pool_size=pool_size,           #池化核大小
            pool_type=pool_type,           #池化类型，默认是最大池化
            pool_stride=pool_stride,       #池化步长
            pool_padding=pool_padding      #填充大小
            )

    def forward(self, inputs):
        x = inputs
        for conv in self._conv2d_list:
            x = conv(x)
        x = self._pool2d(x)
        return x

Please complete the definition of VGG network :

class VGGNet(fluid.dygraph.Layer):
    '''
    VGG网络
    '''
    def __init__(self):
        super(VGGNet, self).__init__()
       
        

    def forward(self, inputs, label=None):
        """前向计算"""
        
        

Fourth, the model training

all_train_iter=0
all_train_iters=[]
all_train_costs=[]
all_train_accs=[]

def draw_train_process(title,iters,costs,accs,label_cost,lable_acc):
    plt.title(title, fontsize=24)
    plt.xlabel("iter", fontsize=20)
    plt.ylabel("cost/acc", fontsize=20)
    plt.plot(iters, costs,color='red',label=label_cost) 
    plt.plot(iters, accs,color='green',label=lable_acc) 
    plt.legend()
    plt.grid()
    plt.show()


def draw_process(title,color,iters,data,label):
    plt.title(title, fontsize=24)
    plt.xlabel("iter", fontsize=20)
    plt.ylabel(label, fontsize=20)
    plt.plot(iters, data,color=color,label=label) 
    plt.legend()
    plt.grid()
    plt.show()

'''
模型训练
'''
# with fluid.dygraph.guard(place = fluid.CUDAPlace(0)):
with fluid.dygraph.guard():
    print(train_parameters['class_dim'])
    print(train_parameters['label_dict'])
    vgg = VGGNet()
    optimizer=fluid.optimizer.AdamOptimizer(learning_rate=train_parameters['learning_strategy']['lr'],parameter_list=vgg.parameters()) 
    for epoch_num in range(train_parameters['num_epochs']):
        for batch_id, data in enumerate(train_reader()):
            dy_x_data = np.array([x[0] for x in data]).astype('float32')           
            y_data = np.array([x[1] for x in data]).astype('int64')      
            y_data = y_data[:, np.newaxis]

            #将Numpy转换为DyGraph接收的输入
            img = fluid.dygraph.to_variable(dy_x_data)
            label = fluid.dygraph.to_variable(y_data)

            out,acc = vgg(img,label)
            loss = fluid.layers.cross_entropy(out, label)
            avg_loss = fluid.layers.mean(loss)

            #使用backward()方法可以执行反向网络
            avg_loss.backward()
            optimizer.minimize(avg_loss)
             
            #将参数梯度清零以保证下一轮训练的正确性
            vgg.clear_gradients()
            

            all_train_iter=all_train_iter+train_parameters['train_batch_size']
            all_train_iters.append(all_train_iter)
            all_train_costs.append(loss.numpy()[0])
            all_train_accs.append(acc.numpy()[0])
                
            if batch_id % 1 == 0:
                print("Loss at epoch {} step {}: {}, acc: {}".format(epoch_num, batch_id, avg_loss.numpy(), acc.numpy()))

    draw_train_process("training",all_train_iters,all_train_costs,all_train_accs,"trainning cost","trainning acc")  
    draw_process("trainning loss","red",all_train_iters,all_train_costs,"trainning loss")
    draw_process("trainning acc","green",all_train_iters,all_train_accs,"trainning acc")  
    
    #保存模型参数
    fluid.save_dygraph(vgg.state_dict(), "vgg")   
    print("Final loss: {}".format(avg_loss.numpy()))

Fifth, model validation

'''
模型校验
'''
with fluid.dygraph.guard():
    model, _ = fluid.load_dygraph("vgg")
    vgg = VGGNet()
    vgg.load_dict(model)
    vgg.eval()
    accs = []
    for batch_id, data in enumerate(eval_reader()):
        dy_x_data = np.array([x[0] for x in data]).astype('float32')
        y_data = np.array([x[1] for x in data]).astype('int')
        y_data = y_data[:, np.newaxis]
        
        img = fluid.dygraph.to_variable(dy_x_data)
        label = fluid.dygraph.to_variable(y_data)

        out, acc = vgg(img, label)
        lab = np.argsort(out.numpy())
        accs.append(acc.numpy()[0])
print(np.mean(accs))

Sixth, model predictions

def load_image(img_path):
    '''
    预测图片预处理
    '''
    img = Image.open(img_path) 
    if img.mode != 'RGB': 
        img = img.convert('RGB') 
    img = img.resize((224, 224), Image.BILINEAR)
    img = np.array(img).astype('float32') 
    img = img.transpose((2, 0, 1))  # HWC to CHW 
    img = img/255                # 像素值归一化 
    return img

label_dic = train_parameters['label_dict']

'''
模型预测
'''with fluid.dygraph.guard():
    model, _ = fluid.dygraph.load_dygraph("vgg")
    vgg = VGGNet()
    vgg.load_dict(model)
    vgg.eval()
    
    #展示预测图片
    infer_path='/home/aistudio/data/data23615/infer_mask01.jpg'
    img = Image.open(infer_path)
    plt.imshow(img)          #根据数组绘制图像
    plt.show()               #显示图像

    #对预测图片进行预处理
    infer_imgs = []
    infer_imgs.append(load_image(infer_path))
    infer_imgs = np.array(infer_imgs)
   
    for  i in range(len(infer_imgs)):
        data = infer_imgs[i]
        dy_x_data = np.array(data).astype('float32')
        dy_x_data=dy_x_data[np.newaxis,:, : ,:]
        img = fluid.dygraph.to_variable(dy_x_data)
        out = vgg(img)
        lab = np.argmax(out.numpy())  #argmax():返回最大数的索引
        print("第{}个样本,被预测为：{}".format(i+1,label_dic[str(lab)]))
        
print("结束")

finish homework

VGG defined network:

A few days ago, and the code is similar, but today many of the sample code throughout the model parameters were unified package deal, and added a ConvPool class, the convolution compounds and pool together, that would be more convenient, we also it's that use.

class VGGNet(fluid.dygraph.Layer):
    '''
    VGG网络
    '''
    def __init__(self):
        super(VGGNet, self).__init__()
        # 通道数、卷积核个数、卷积核大小、池化核大小、池化步长、连续卷积个数
        self.convpool01 = ConvPool(3, 64, 3, 2, 2, 2, act='relu')
        self.convpool02 = ConvPool(64, 128, 3, 2, 2, 2, act='relu')
        self.convpool03 = ConvPool(128, 256, 3, 2, 2, 3, act='relu')
        self.convpool04 = ConvPool(256, 512, 3, 2, 2, 3, act='relu')
        self.convpool05 = ConvPool(512, 512, 3, 2, 2, 3, act='relu')

        self.pool_5_shape = 512*7*7
        self.fc01 = fluid.dygraph.Linear(self.pool_5_shape, 4096, act='relu')
        self.fc02 = fluid.dygraph.Linear(4096, 4096, act='relu')
        self.fc03 = fluid.dygraph.Linear(4096, 2, act='softmax')

    def forward(self, inputs, label=None):
        """前向计算"""
        out = self.convpool01(inputs)
        out = self.convpool02(out)
        out = self.convpool03(out)
        out = self.convpool04(out)
        out = self.convpool05(out)
        
        out = fluid.layers.reshape(out, shape=[-1, 512*7*7])
        out = self.fc01(out)
        out = self.fc02(out)
        out = self.fc03(out)

        if label is not None:
        	acc = fluid.layers.accuracy(input=out, label=label)
        	return out, acc
        else:
        	return out

Number of training wheels of our initial take is 10, you can see the accuracy of the model train is the ups and downs.

Drawn out of the picture, too, the accuracy of the model has been trained in shock.

The accuracy of the test set at about 0.6.

Masks recognition is a binary classification, the results only wear masks and not wearing masks two kinds. We predict Shihai masks picture barely able to predict success.

Accuracy rate is only about 0.6 think we certainly still have to go optimized parameters were transferred from the following three aspects.

• Several rounds of training, that is, the number of iterations (num_epochs)
• Learning rate (learningrate)
• Each batch size (batch_size) training

We increased the number of training rounds, increased to 20 times, cut the learning rate, fell to 0.0001, increasing the size of each batch of training, increased to 16. These parameters can be modified environment configuration of the first step.

Emphasize here that a slightly increase in the batch size can be trained to improve the accuracy, but batch_size not just the size of the transfer, generally a multiple of 8, the highest efficiency of such parallel operation inside the GPU .

After training you can see, the accuracy of the training set gradually converge to 1.0.

The accuracy of the test set also reached 1.0, very nice.

Because the data is relatively small study, generalization ability is not strong, the group of bigwigs feedback that can be suitably used data enhancement (Data Augmentation) methods to improve the prediction of a real scene success rate.