目录
禁止转载，侵权必究
前言
我们上一章介绍了MNIST手写数字识别的问题的解法。一步步从单层线性网络升级为多层CNN网络，并且介绍了配套的常用函数，比如relu，softmax，cross_entropy。最后我们还评估了CNN网络的计算量，并且计算了每一层网络的输入、参数、输出的形状。
我们上一章教程介绍的CNN网络叫做LeNet，那么机器学习是不是“一招鲜吃遍天”呢？我们来看看各种CNN算法在iChallenge-PM问题中的表现。
识别目标
iChallenge-PM是百度大脑和中山大学中山眼科中心联合举办的iChallenge比赛中，提供的关于病理性近视（Pathologic Myopia，PM）的医疗类数据集，包含1200个受试者的眼底视网膜图片，训练、验证和测试数据集各400张。
公用代码
import cv2
import random
import numpy as np
# 对读入的图像数据进行预处理
def transform_img(img):
# 将图片尺寸缩放道 224x224
img = cv2.resize(img, (224, 224))
# 读入的图像数据格式是[H, W, C]
# 使用转置操作将其变成[C, H, W]
img = np.transpose(img, (2,0,1))
img = img.astype('float32')
# 将数据范围调整到[-1.0, 1.0]之间
img = img / 255.
img = img * 2.0 - 1.0
return img
定义训练集数据读取器
# 定义训练集数据读取器
def data_loader(datadir, batch_size=10, mode = 'train'):
# 将datadir目录下的文件列出来，每条文件都要读入
filenames = os.listdir(datadir)
def reader():
if mode == 'train':
# 训练时随机打乱数据顺序
random.shuffle(filenames)
batch_imgs = []
batch_labels = []
for name in filenames:
filepath = os.path.join(datadir, name)
img = cv2.imread(filepath)
img = transform_img(img)
if name[0] == 'H' or name[0] == 'N':
# H开头的文件名表示高度近似，N开头的文件名表示正常视力
# 高度近视和正常视力的样本，都不是病理性的，属于负样本，标签为0
label = 0
elif name[0] == 'P':
# P开头的是病理性近视，属于正样本，标签为1
label = 1
else:
raise('Not excepted file name')
# 每读取一个样本的数据，就将其放入数据列表中
batch_imgs.append(img)
batch_labels.append(label)
if len(batch_imgs) == batch_size:
# 当数据列表的长度等于batch_size的时候，
# 把这些数据当作一个mini-batch，并作为数据生成器的一个输出
imgs_array = np.array(batch_imgs).astype('float32')
labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
yield imgs_array, labels_array
batch_imgs = []
batch_labels = []
if len(batch_imgs) > 0:
# 剩余样本数目不足一个batch_size的数据，一起打包成一个mini-batch
imgs_array = np.array(batch_imgs).astype('float32')
labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
yield imgs_array, labels_array
return reader定义验证集数据读取器
# 定义验证集数据读取器
def valid_data_loader(datadir, csvfile, batch_size=10, mode='valid'):
# 训练集读取时通过文件名来确定样本标签，验证集则通过csvfile来读取每个图片对应的标签
# 请查看解压后的验证集标签数据，观察csvfile文件里面所包含的内容
# csvfile文件所包含的内容格式如下，每一行代表一个样本，
# 其中第一列是图片id，第二列是文件名，第三列是图片标签，
# 第四列和第五列是Fovea的坐标，与分类任务无关
# ID,imgName,Label,Fovea_X,Fovea_Y
# 1,V0001.jpg,0,1157.74,1019.87
# 2,V0002.jpg,1,1285.82,1080.47
# 打开包含验证集标签的csvfile，并读入其中的内容
filelists = open(csvfile).readlines()
def reader():
batch_imgs = []
batch_labels = []
for line in filelists[1:]:
line = line.strip().split(',')
name = line[1]
label = int(line[2])
# 根据图片文件名加载图片，并对图像数据作预处理
filepath = os.path.join(datadir, name)
img = cv2.imread(filepath)
img = transform_img(img)
# 每读取一个样本的数据，就将其放入数据列表中
batch_imgs.append(img)
batch_labels.append(label)
if len(batch_imgs) == batch_size:
# 当数据列表的长度等于batch_size的时候，
# 把这些数据当作一个mini-batch，并作为数据生成器的一个输出
imgs_array = np.array(batch_imgs).astype('float32')
labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
yield imgs_array, labels_array
batch_imgs = []
batch_labels = []
if len(batch_imgs) > 0:
# 剩余样本数目不足一个batch_size的数据，一起打包成一个mini-batch
imgs_array = np.array(batch_imgs).astype('float32')
labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
yield imgs_array, labels_array
return reader数据形状
# 查看数据形状
DATADIR = '/home/aistudio/work/palm/PALM-Training400/PALM-Training400'
train_loader = data_loader(DATADIR,
batch_size=10, mode='train')
data_reader = train_loader()
data = next(data_reader)
data[0].shape, data[1].shape我们可以看到结果：
((10, 3, 224, 224), (10, 1))
每个batch有10张图，3个颜色通道，224×224的尺寸。标签label有10个，维度为1。
经典CNN网络-LeNet
1.import依赖
import os
import random
import paddle
import paddle.fluid as fluid
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, Linear
import numpy as np
# 百度AI Studio数据文件路径
DATADIR = '/home/aistudio/work/palm/PALM-Training400/PALM-Training400'
DATADIR2 = '/home/aistudio/work/palm/PALM-Validation400'
CSVFILE = '/home/aistudio/labels.csv'2.定义模型
# 定义 LeNet 网络结构
class LeNet(fluid.dygraph.Layer):
def __init__(self, num_classes=1):
super(LeNet, self).__init__()
# 创建卷积和池化层块，每个卷积层使用Sigmoid激活函数，后面跟着一个2x2的池化
self.conv1 = Conv2D(num_channels=3, num_filters=6, filter_size=5, act='sigmoid')
self.pool1 = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
self.conv2 = Conv2D(num_channels=6, num_filters=16, filter_size=5, act='sigmoid')
self.pool2 = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
# 创建第3个卷积层
self.conv3 = Conv2D(num_channels=16, num_filters=120, filter_size=4, act='sigmoid')
# 创建全连接层，第一个全连接层的输出神经元个数为64， 第二个全连接层输出神经元个数为分类标签的类别数
self.fc1 = Linear(input_dim=300000, output_dim=64, act='sigmoid')
self.fc2 = Linear(input_dim=64, output_dim=num_classes)
# 网络的前向计算过程
def forward(self, x):
x = self.conv1(x)
x = self.pool1(x)
x = self.conv2(x)
x = self.pool2(x)
x = self.conv3(x)
x = fluid.layers.reshape(x, [x.shape[0], -1])
x = self.fc1(x)
x = self.fc2(x)
return x3.训练过程
# 定义训练过程
def train(model):
with fluid.dygraph.guard():
print('start training ... ')
model.train()
epoch_num = 5
# 定义优化器
opt = fluid.optimizer.Momentum(learning_rate=0.001, momentum=0.9, parameter_list=model.parameters())
# 定义数据读取器，训练数据读取器和验证数据读取器
train_loader = data_loader(DATADIR, batch_size=10, mode='train')
valid_loader = valid_data_loader(DATADIR2, CSVFILE)
for epoch in range(epoch_num):
for batch_id, data in enumerate(train_loader()):
x_data, y_data = data
img = fluid.dygraph.to_variable(x_data)
label = fluid.dygraph.to_variable(y_data)
# 运行模型前向计算，得到预测值
logits = model(img)
# 进行loss计算
loss = fluid.layers.sigmoid_cross_entropy_with_logits(logits, label)
avg_loss = fluid.layers.mean(loss)
if batch_id % 10 == 0:
print("epoch: {}, batch_id: {}, loss is: {}".format(epoch, batch_id, avg_loss.numpy()))
# 反向传播，更新权重，清除梯度
avg_loss.backward()
opt.minimize(avg_loss)
model.clear_gradients()
model.eval()
accuracies = []
losses = []
for batch_id, data in enumerate(valid_loader()):
x_data, y_data = data
img = fluid.dygraph.to_variable(x_data)
label = fluid.dygraph.to_variable(y_data)
# 运行模型前向计算，得到预测值
logits = model(img)
# 二分类，sigmoid计算后的结果以0.5为阈值分两个类别
# 计算sigmoid后的预测概率，进行loss计算
pred = fluid.layers.sigmoid(logits)
loss = fluid.layers.sigmoid_cross_entropy_with_logits(logits, label)
# 计算预测概率小于0.5的类别
pred2 = pred * (-1.0) + 1.0
# 得到两个类别的预测概率，并沿第一个维度级联
pred = fluid.layers.concat([pred2, pred], axis=1)
acc = fluid.layers.accuracy(pred, fluid.layers.cast(label, dtype='int64'))
accuracies.append(acc.numpy())
losses.append(loss.numpy())
print("[validation] accuracy/loss: {}/{}".format(np.mean(accuracies), np.mean(losses)))
model.train()
# save params of model
fluid.save_dygraph(model.state_dict(), 'mnist')
# save optimizer state
fluid.save_dygraph(opt.state_dict(), 'mnist')
3.评估过程
# 定义评估过程
def evaluation(model, params_file_path):
with fluid.dygraph.guard():
print('start evaluation .......')
#加载模型参数
model_state_dict, _ = fluid.load_dygraph(params_file_path)
model.load_dict(model_state_dict)
model.eval()
eval_loader = load_data('eval')
acc_set = []
avg_loss_set = []
for batch_id, data in enumerate(eval_loader()):
x_data, y_data = data
img = fluid.dygraph.to_variable(x_data)
label = fluid.dygraph.to_variable(y_data)
# 计算预测和精度
prediction, acc = model(img, label)
# 计算损失函数值
loss = fluid.layers.cross_entropy(input=prediction, label=label)
avg_loss = fluid.layers.mean(loss)
acc_set.append(float(acc.numpy()))
avg_loss_set.append(float(avg_loss.numpy()))
# 求平均精度
acc_val_mean = np.array(acc_set).mean()
avg_loss_val_mean = np.array(avg_loss_set).mean()
print('loss={}, acc={}'.format(avg_loss_val_mean, acc_val_mean))4.查看训练和评估的结果
start training ...
epoch: 0, batch_id: 0, loss is: [0.78380126]
epoch: 0, batch_id: 10, loss is: [0.7035402]
epoch: 0, batch_id: 20, loss is: [0.69243866]
epoch: 0, batch_id: 30, loss is: [0.6760834]
[validation] accuracy/loss: 0.5275000333786011/0.6938738822937012
epoch: 1, batch_id: 0, loss is: [0.6761131]
epoch: 1, batch_id: 10, loss is: [0.6756402]
epoch: 1, batch_id: 20, loss is: [0.67702425]
epoch: 1, batch_id: 30, loss is: [0.6707235]
[validation] accuracy/loss: 0.5275000333786011/0.6917563080787659
epoch: 2, batch_id: 0, loss is: [0.69567424]
epoch: 2, batch_id: 10, loss is: [0.6799297]
epoch: 2, batch_id: 20, loss is: [0.6792191]
epoch: 2, batch_id: 30, loss is: [0.6964021]
[validation] accuracy/loss: 0.5275000333786011/0.691692054271698
epoch: 3, batch_id: 0, loss is: [0.7113117]
epoch: 3, batch_id: 10, loss is: [0.69906384]
epoch: 3, batch_id: 20, loss is: [0.69818497]
epoch: 3, batch_id: 30, loss is: [0.67282254]
[validation] accuracy/loss: 0.5275000333786011/0.6927729249000549
epoch: 4, batch_id: 0, loss is: [0.6572024]
epoch: 4, batch_id: 10, loss is: [0.65163314]
epoch: 4, batch_id: 20, loss is: [0.702415]
epoch: 4, batch_id: 30, loss is: [0.7255191]
[validation] accuracy/loss: 0.5275000333786011/0.691861629486084结论：LeNet并不能解决iChallenge-PM问题。
CNN网络-AlexNet
1.定义AlexNet网络
# -*- coding:utf-8 -*-
# 导入需要的包
import paddle
import paddle.fluid as fluid
import numpy as np
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, Linear
# 定义 AlexNet 网络结构
class AlexNet(fluid.dygraph.Layer):
def __init__(self, num_classes=1):
super(AlexNet, self).__init__()
# AlexNet与LeNet一样也会同时使用卷积和池化层提取图像特征
# 与LeNet不同的是激活函数换成了‘relu’
self.conv1 = Conv2D(num_channels=3, num_filters=96, filter_size=11, stride=4, padding=5, act='relu')
self.pool1 = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
self.conv2 = Conv2D(num_channels=96, num_filters=256, filter_size=5, stride=1, padding=2, act='relu')
self.pool2 = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
self.conv3 = Conv2D(num_channels=256, num_filters=384, filter_size=3, stride=1, padding=1, act='relu')
self.conv4 = Conv2D(num_channels=384, num_filters=384, filter_size=3, stride=1, padding=1, act='relu')
self.conv5 = Conv2D(num_channels=384, num_filters=256, filter_size=3, stride=1, padding=1, act='relu')
self.pool5 = Pool2D(pool_size=2, pool_stride=2, pool_type='max')
self.fc1 = Linear(input_dim=12544, output_dim=4096, act='relu')
self.drop_ratio1 = 0.5
self.fc2 = Linear(input_dim=4096, output_dim=4096, act='relu')
self.drop_ratio2 = 0.5
self.fc3 = Linear(input_dim=4096, output_dim=num_classes)
def forward(self, x):
x = self.conv1(x)
x = self.pool1(x)
x = self.conv2(x)
x = self.pool2(x)
x = self.conv3(x)
x = self.conv4(x)
x = self.conv5(x)
x = self.pool5(x)
x = fluid.layers.reshape(x, [x.shape[0], -1])
x = self.fc1(x)
# 在全连接之后使用dropout抑制过拟合
x= fluid.layers.dropout(x, self.drop_ratio1)
x = self.fc2(x)
# 在全连接之后使用dropout抑制过拟合
x = fluid.layers.dropout(x, self.drop_ratio2)
x = self.fc3(x)
return x2. 复用LeNet的训练&评估过程
with fluid.dygraph.guard():
model = AlexNet()
train(model)3.查看训练和评估的结果
start training ...
epoch: 0, batch_id: 0, loss is: [0.8983343]
epoch: 0, batch_id: 10, loss is: [0.78235567]
epoch: 0, batch_id: 20, loss is: [0.5539009]
epoch: 0, batch_id: 30, loss is: [0.60531735]
[validation] accuracy/loss: 0.8999999761581421/0.5300776958465576
epoch: 1, batch_id: 0, loss is: [0.56365806]
epoch: 1, batch_id: 10, loss is: [0.435232]
epoch: 1, batch_id: 20, loss is: [0.37820452]
epoch: 1, batch_id: 30, loss is: [0.34011936]
[validation] accuracy/loss: 0.9325000047683716/0.25468358397483826
epoch: 2, batch_id: 0, loss is: [0.32455772]
epoch: 2, batch_id: 10, loss is: [0.5270335]
epoch: 2, batch_id: 20, loss is: [0.41049075]
epoch: 2, batch_id: 30, loss is: [0.14482744]
[validation] accuracy/loss: 0.9175001382827759/0.25097012519836426
epoch: 3, batch_id: 0, loss is: [0.43380293]
epoch: 3, batch_id: 10, loss is: [0.12313936]
epoch: 3, batch_id: 20, loss is: [0.5423292]
epoch: 3, batch_id: 30, loss is: [0.7778549]
[validation] accuracy/loss: 0.9049999117851257/0.28984639048576355
epoch: 4, batch_id: 0, loss is: [0.14641385]
epoch: 4, batch_id: 10, loss is: [0.5889987]
epoch: 4, batch_id: 20, loss is: [0.0941387]
epoch: 4, batch_id: 30, loss is: [0.20460884]
[validation] accuracy/loss: 0.9149999618530273/0.2128506302833557结论：AlexNet可以解决iChallenge-PM问题。
CNN网络-VGG
1.import依赖包
import numpy as np
import paddle
import paddle.fluid as fluid
from paddle.fluid.layer_helper import LayerHelper
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, BatchNorm, Linear
from paddle.fluid.dygraph.base import to_variable1.定义VGG网络
因为VGG网络是深度神经网络，我们把重复的部分作为一个block单独定义。总共定义5个block，第一个block有2个卷积层和1个池化层，第二个block有2个卷积层和1个池化层，第三个block有3个卷积层和1个池化层，第四个block有3个卷积层和1个池化层，第五个block有3个卷积层和1个池化层。最后还有3个全连接层。因此：VGG总共有13层卷积+3层全连接，是典型的深度神经网络。
# 定义vgg块，包含多层卷积和1层2x2的最大池化层
class vgg_block(fluid.dygraph.Layer):
def __init__(self, num_convs, in_channels, out_channels):
"""
num_convs, 卷积层的数目
num_channels, 卷积层的输出通道数，在同一个Incepition块内，卷积层输出通道数是一样的
"""
super(vgg_block, self).__init__()
self.conv_list = []
for i in range(num_convs):
conv_layer = self.add_sublayer('conv_' + str(i), Conv2D(num_channels=in_channels,
num_filters=out_channels, filter_size=3, padding=1, act='relu'))
self.conv_list.append(conv_layer)
in_channels = out_channels
self.pool = Pool2D(pool_stride=2, pool_size = 2, pool_type='max')
def forward(self, x):
for item in self.conv_list:
x = item(x)
return self.pool(x)class VGG(fluid.dygraph.Layer):
def __init__(self, conv_arch=((2, 64),
(2, 128), (3, 256), (3, 512), (3, 512))):
super(VGG, self).__init__()
self.vgg_blocks=[]
iter_id = 0
# 添加vgg_block
# 这里一共5个vgg_block，每个block里面的卷积层数目和输出通道数由conv_arch指定
in_channels = [3, 64, 128, 256, 512, 512]
for (num_convs, num_channels) in conv_arch:
block = self.add_sublayer('block_' + str(iter_id),
vgg_block(num_convs, in_channels=in_channels[iter_id],
out_channels=num_channels))
self.vgg_blocks.append(block)
iter_id += 1
self.fc1 = Linear(input_dim=512*7*7, output_dim=4096,
act='relu')
self.drop1_ratio = 0.5
self.fc2= Linear(input_dim=4096, output_dim=4096,
act='relu')
self.drop2_ratio = 0.5
self.fc3 = Linear(input_dim=4096, output_dim=1)
def forward(self, x):
for item in self.vgg_blocks:
x = item(x)
x = fluid.layers.reshape(x, [x.shape[0], -1])
x = fluid.layers.dropout(self.fc1(x), self.drop1_ratio)
x = fluid.layers.dropout(self.fc2(x), self.drop2_ratio)
x = self.fc3(x)
return x2. 复用LeNet的训练&评估过程
with fluid.dygraph.guard():
model = VGG()
train(model)3.查看训练和评估的结果
start training ...
epoch: 0, batch_id: 0, loss is: [0.77041894]
epoch: 0, batch_id: 10, loss is: [0.68711454]
epoch: 0, batch_id: 20, loss is: [0.68020743]
epoch: 0, batch_id: 30, loss is: [0.53936696]
[validation] accuracy/loss: 0.5300000309944153/0.547136664390564
epoch: 1, batch_id: 0, loss is: [0.623295]
epoch: 1, batch_id: 10, loss is: [0.6252886]
epoch: 1, batch_id: 20, loss is: [0.27016944]
epoch: 1, batch_id: 30, loss is: [0.72312546]
[validation] accuracy/loss: 0.92249995470047/0.2480151355266571
epoch: 2, batch_id: 0, loss is: [0.19115444]
epoch: 2, batch_id: 10, loss is: [0.22194262]
epoch: 2, batch_id: 20, loss is: [0.33472323]
epoch: 2, batch_id: 30, loss is: [0.59723985]
[validation] accuracy/loss: 0.925000011920929/0.22546321153640747
epoch: 3, batch_id: 0, loss is: [0.5138165]
epoch: 3, batch_id: 10, loss is: [0.38347217]
epoch: 3, batch_id: 20, loss is: [0.6232009]
epoch: 3, batch_id: 30, loss is: [0.34507117]
[validation] accuracy/loss: 0.8949999809265137/0.305342435836792
epoch: 4, batch_id: 0, loss is: [0.21837215]
epoch: 4, batch_id: 10, loss is: [0.5486041]
epoch: 4, batch_id: 20, loss is: [0.488666]
epoch: 4, batch_id: 30, loss is: [0.27352032]
[validation] accuracy/loss: 0.9274999499320984/0.21110011637210846结论：VGG也可以解决iChallenge-PM问题。
评估VGG-16的算力消耗
| 层 | Cin | 输出形状 | 乘法次数 |
| block1_conv0 | 3 | [10,64,224,224] | 约8.67亿 |
| block1_conv1 | 64 | [10,64,224,224] | 约185亿 |
| pooling1 | 64 | [10,64,112,112] | – |
| block2_conv0 | 64 | [10,128,112,112] | 约92.5亿 |
| block2_conv1 | 128 | [10,128,112,112] | 约185亿 |
| pooling2 | 128 | [10,128,56,56] | – |
| block3_conv0 | 128 | [10,256,56,56] | 约92.5亿 |
| block3_conv1 | 256 | [10,256,56,56] | 约185亿 |
| block3_conv2 | 256 | [10,256,56,56] | 约185亿 |
| pooling3 | 256 | [10,256,28,28] | – |
| block4_conv0 | 256 | [10,512,28,28] | 约92.5亿 |
| block4_conv1 | 512 | [10,512,28,28] | 约185亿 |
| block4_conv2 | 512 | [10,512,28,28] | 约185亿 |
| pooling4 | 512 | [10,512,14,14] | – |
| block5_conv0 | 512 | [10,512,14,14] | 约46.2亿 |
| block5_conv1 | 512 | [10,512,14,14] | 约46.2亿 |
| block5_conv2 | 512 | [10,512,14,14] | 约46.2亿 |
| pooling5 | 512 | [10,512,7,7] | – |
| fc1 | 25088 | [10,4096] | – |
| fc2 | 4096 | [10,4096] | – |
| fc3 | 4096 | [10,1] | – |
所以本例中VGG-16每一个批次的乘法数约1534.77亿。因为加法和乘法算力消耗相当，所以总算力消耗为3069.54亿次。