第P7周：马铃薯病害识别
- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍖 原作者：K同学啊
-
要求：
- 自己搭建VGG-16网络框架
- 调用官方的VGG-16网络框架
- 如何查看模型的参数量以及相关指标
-
拔高（可选）：
- 验证集准确率达到100%
- 使用PPT画出VGG-16算法框架图（发论文需要这项技能）
-
探索（难度有点大）
- 在不影响准确率的前提下轻量化模型
- 目前VGG16的Total params是134,272,835
-
我的环境：
- 操作系统：CentOS7
- 显卡：RTX3090
- 显卡驱动：535.154.05
- CUDA版本: 12.2
- 语言环境：Python3.10
- 编译器：Jupyter Lab
- 深度学习环境：
- torch==12.1
- torchvision==0.18.1
这次我们使用的是马铃薯病害数据集，该数据集包含表现出各种疾病的马铃薯植物的高分辨率图像，包括早期疫病、晚期疫病和健康叶子。它旨在帮助开发和测试图像识别模型，以实现准确的疾病检测和分类，从而促进农业诊断的进步。
一、前期准备
1. 设置GPU
import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import os,PIL,pathlib,warnings
warnings.filterwarnings("ignore") #忽略警告信息
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
device
device(type='cuda')
2. 导入数据
import os,PIL,random,pathlib
data_dir = './data/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("/")[1] for path in data_paths]
classeNames
['healthy', 'Early_blight', 'Late_blight']
# 关于transforms.Compose的更多介绍可以参考：https://blog.csdn.net/qq_38251616/article/details/124878863
train_transforms = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
transforms.RandomHorizontalFlip(), # 随机水平翻转
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
test_transform = transforms.Compose([
transforms.Resize([224, 224]), # 将输入图片resize成统一尺寸
transforms.RandomHorizontalFlip(), # 随机水平翻转
transforms.ToTensor(), # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
transforms.Normalize( # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])
total_data = datasets.ImageFolder("./data/",transform=train_transforms)
total_data
Dataset ImageFolder
Number of datapoints: 2152
Root location: ./data/
StandardTransform
Transform: Compose(
Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=True)
RandomHorizontalFlip(p=0.5)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)
total_data.class_to_idx
{'Early_blight': 0, 'Late_blight': 1, 'healthy': 2}
3. 划分数据集
train_size = int(0.8 * len(total_data))
test_size = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
train_dataset, test_dataset
(<torch.utils.data.dataset.Subset at 0x7fa5eb6e93c0>,
<torch.utils.data.dataset.Subset at 0x7fa5eb6ebdc0>)
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=1)
for X, y in test_dl:
print("Shape of X [N, C, H, W]: ", X.shape)
print("Shape of y: ", y.shape, y.dtype)
break
Shape of X [N, C, H, W]: torch.Size([32, 3, 224, 224])
Shape of y: torch.Size([32]) torch.int64
二、手动搭建VGG-16模型
1. 模型介绍
- VVG-16结构说明：
- 13个卷积层（Convolutional Layer），分别用blockX_convX表示
- 3个全连接层（Fully connected Layer），分别用fcX与predictions表示
- 5个池化层（Pool layer），分别用blockX_pool表示
VGG-16包含了16个隐藏层（13个卷积层和3个全连接层），故称为VGG-16
2. 模型搭建
import torch.nn.functional as F
class vgg16(nn.Module):
def __init__(self):
super(vgg16, self).__init__()
# 卷积块1
self.block1 = nn.Sequential(
nn.Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块2
self.block2 = nn.Sequential(
nn.Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块3
self.block3 = nn.Sequential(
nn.Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块4
self.block4 = nn.Sequential(
nn.Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块5
self.block5 = nn.Sequential(
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 全连接网络层，用于分类
self.classifier = nn.Sequential(
nn.Linear(in_features=512*7*7, out_features=4096),
nn.ReLU(),
nn.Linear(in_features=4096, out_features=4096),
nn.ReLU(),
nn.Linear(in_features=4096, out_features=3)
)
def forward(self, x):
x = self.block1(x)
x = self.block2(x)
x = self.block3(x)
x = self.block4(x)
x = self.block5(x)
x = torch.flatten(x, start_dim=1)
x = self.classifier(x)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
model = vgg16().to(device)
model
Using cuda device
vgg16(
(block1): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block2): Sequential(
(0): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block3): Sequential(
(0): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block4): Sequential(
(0): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block5): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(classifier): Sequential(
(0): Linear(in_features=25088, out_features=4096, bias=True)
(1): ReLU()
(2): Linear(in_features=4096, out_features=4096, bias=True)
(3): ReLU()
(4): Linear(in_features=4096, out_features=3, bias=True)
)
)
3. 查看模型详情
import torchsummary as summary
summary.summary(model, input_size=(3, 224, 224))
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 224, 224] 1,792
ReLU-2 [-1, 64, 224, 224] 0
Conv2d-3 [-1, 64, 224, 224] 36,928
ReLU-4 [-1, 64, 224, 224] 0
MaxPool2d-5 [-1, 64, 112, 112] 0
Conv2d-6 [-1, 128, 112, 112] 73,856
ReLU-7 [-1, 128, 112, 112] 0
Conv2d-8 [-1, 128, 112, 112] 147,584
ReLU-9 [-1, 128, 112, 112] 0
MaxPool2d-10 [-1, 128, 56, 56] 0
Conv2d-11 [-1, 256, 56, 56] 295,168
ReLU-12 [-1, 256, 56, 56] 0
Conv2d-13 [-1, 256, 56, 56] 590,080
ReLU-14 [-1, 256, 56, 56] 0
Conv2d-15 [-1, 256, 56, 56] 590,080
ReLU-16 [-1, 256, 56, 56] 0
MaxPool2d-17 [-1, 256, 28, 28] 0
Conv2d-18 [-1, 512, 28, 28] 1,180,160
ReLU-19 [-1, 512, 28, 28] 0
Conv2d-20 [-1, 512, 28, 28] 2,359,808
ReLU-21 [-1, 512, 28, 28] 0
Conv2d-22 [-1, 512, 28, 28] 2,359,808
ReLU-23 [-1, 512, 28, 28] 0
MaxPool2d-24 [-1, 512, 14, 14] 0
Conv2d-25 [-1, 512, 14, 14] 2,359,808
ReLU-26 [-1, 512, 14, 14] 0
Conv2d-27 [-1, 512, 14, 14] 2,359,808
ReLU-28 [-1, 512, 14, 14] 0
Conv2d-29 [-1, 512, 14, 14] 2,359,808
ReLU-30 [-1, 512, 14, 14] 0
MaxPool2d-31 [-1, 512, 7, 7] 0
Linear-32 [-1, 4096] 102,764,544
ReLU-33 [-1, 4096] 0
Linear-34 [-1, 4096] 16,781,312
ReLU-35 [-1, 4096] 0
Linear-36 [-1, 3] 12,291
================================================================
Total params: 134,272,835
Trainable params: 134,272,835
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 218.52
Params size (MB): 512.21
Estimated Total Size (MB): 731.30
----------------------------------------------------------------
4. 自建VGG-16模型
import torch.nn.functional as F
class VGG16Model(nn.Module):
def __init__(self, nclasses=10):
super(VGG16Model, self).__init__()
self.conv11=nn.Sequential(
nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1), ## 224*224*64
nn.BatchNorm2d(64),
nn.ReLU())
self.conv12=nn.Sequential(
nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1), ## 224*224*64
nn.BatchNorm2d(64),
nn.ReLU())
self.pool1=nn.Sequential(
nn.MaxPool2d(2)) ## 112*112*64
self.conv21=nn.Sequential(
nn.Conv2d(64, 128, kernel_size=3, padding=1), ## 112*112*128
nn.BatchNorm2d(128),
nn.ReLU())
self.conv22=nn.Sequential(
nn.Conv2d(128, 128, kernel_size=3, padding=1), ## 112*112*128
nn.BatchNorm2d(128),
nn.ReLU())
self.pool2=nn.Sequential(
nn.MaxPool2d(2)) ## 56*56*128
self.conv31=nn.Sequential(
nn.Conv2d(128, 256, kernel_size=3, padding=1), ## 56*56*256
nn.BatchNorm2d(256),
nn.ReLU())
self.conv32=nn.Sequential(
nn.Conv2d(256, 256, kernel_size=3, padding=1), ## 56*56*256
nn.BatchNorm2d(256),
nn.ReLU())
self.conv33=nn.Sequential(
nn.Conv2d(256, 256, kernel_size=3, padding=1), ## 56*56*256
nn.BatchNorm2d(256),
nn.ReLU())
self.pool3=nn.Sequential(
nn.MaxPool2d(2)) ## 28*28*256
self.conv41=nn.Sequential(
nn.Conv2d(256, 512, kernel_size=3, padding=1), ## 28*28*512
nn.BatchNorm2d(512),
nn.ReLU())
self.conv42=nn.Sequential(
nn.Conv2d(512, 512, kernel_size=3, padding=1), ## 28*28*512
nn.BatchNorm2d(512),
nn.ReLU())
self.conv43=nn.Sequential(
nn.Conv2d(512, 512, kernel_size=3, padding=1), ## 28*28*512
nn.BatchNorm2d(512),
nn.ReLU())
self.pool4=nn.Sequential(
nn.MaxPool2d(2)) ## 14*14*512
self.conv51=nn.Sequential(
nn.Conv2d(512, 512, kernel_size=3, padding=1), ## 14*14*512
nn.BatchNorm2d(512),
nn.ReLU())
self.conv52=nn.Sequential(
nn.Conv2d(512, 512, kernel_size=3, padding=1), ## 14*14*512
nn.BatchNorm2d(512),
nn.ReLU())
self.conv53=nn.Sequential(
nn.Conv2d(512, 512, kernel_size=3, padding=1), ## 14*14*512
nn.BatchNorm2d(512),
nn.ReLU())
self.pool5=nn.Sequential(
nn.MaxPool2d(2)) ## 7*7*512
self.fc1 = nn.Sequential(
nn.Dropout(0.5), ## 前面vgg16模型没有这一步
nn.Linear(7*7*512, 4096),
nn.ReLU())
self.fc2 = nn.Sequential(
nn.Dropout(0.5), ## 前面vgg16模型没有这一步
nn.Linear(4096, 4096),
nn.ReLU())
self.fc3 = nn.Sequential(
nn.Linear(4096, nclasses))
def forward(self, x):
batch_size = x.size(0)
x = self.conv11(x)
x = self.conv12(x)
x = self.pool1(x)
x = self.conv21(x)
x = self.conv22(x)
x = self.pool2(x)
x = self.conv31(x)
x = self.conv32(x)
x = self.conv33(x)
x = self.pool3(x)
x = self.conv41(x)
x = self.conv42(x)
x = self.conv43(x)
x = self.pool4(x)
x = self.conv51(x)
x = self.conv52(x)
x = self.conv53(x)
x = self.pool5(x)
x = x.view(batch_size, -1)
x = self.fc1(x)
x = self.fc2(x)
x = self.fc3(x)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
#device = "cpu"
print("Using {} device".format(device))
vgg16_model = VGG16Model(len(classeNames)).to(device)
vgg16_model
Using cuda device
VGG16Model(
(conv11): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv12): Sequential(
(0): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool1): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(conv21): Sequential(
(0): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv22): Sequential(
(0): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool2): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(conv31): Sequential(
(0): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv32): Sequential(
(0): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv33): Sequential(
(0): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool3): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(conv41): Sequential(
(0): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv42): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv43): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool4): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(conv51): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv52): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv53): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool5): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(fc1): Sequential(
(0): Dropout(p=0.5, inplace=False)
(1): Linear(in_features=25088, out_features=4096, bias=True)
(2): ReLU()
)
(fc2): Sequential(
(0): Dropout(p=0.5, inplace=False)
(1): Linear(in_features=4096, out_features=4096, bias=True)
(2): ReLU()
)
(fc3): Sequential(
(0): Linear(in_features=4096, out_features=3, bias=True)
)
)
import torchsummary as summary
summary.summary(vgg16_model, (3, 224, 224))
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 224, 224] 1,792
BatchNorm2d-2 [-1, 64, 224, 224] 128
ReLU-3 [-1, 64, 224, 224] 0
Conv2d-4 [-1, 64, 224, 224] 36,928
BatchNorm2d-5 [-1, 64, 224, 224] 128
ReLU-6 [-1, 64, 224, 224] 0
MaxPool2d-7 [-1, 64, 112, 112] 0
Conv2d-8 [-1, 128, 112, 112] 73,856
BatchNorm2d-9 [-1, 128, 112, 112] 256
ReLU-10 [-1, 128, 112, 112] 0
Conv2d-11 [-1, 128, 112, 112] 147,584
BatchNorm2d-12 [-1, 128, 112, 112] 256
ReLU-13 [-1, 128, 112, 112] 0
MaxPool2d-14 [-1, 128, 56, 56] 0
Conv2d-15 [-1, 256, 56, 56] 295,168
BatchNorm2d-16 [-1, 256, 56, 56] 512
ReLU-17 [-1, 256, 56, 56] 0
Conv2d-18 [-1, 256, 56, 56] 590,080
BatchNorm2d-19 [-1, 256, 56, 56] 512
ReLU-20 [-1, 256, 56, 56] 0
Conv2d-21 [-1, 256, 56, 56] 590,080
BatchNorm2d-22 [-1, 256, 56, 56] 512
ReLU-23 [-1, 256, 56, 56] 0
MaxPool2d-24 [-1, 256, 28, 28] 0
Conv2d-25 [-1, 512, 28, 28] 1,180,160
BatchNorm2d-26 [-1, 512, 28, 28] 1,024
ReLU-27 [-1, 512, 28, 28] 0
Conv2d-28 [-1, 512, 28, 28] 2,359,808
BatchNorm2d-29 [-1, 512, 28, 28] 1,024
ReLU-30 [-1, 512, 28, 28] 0
Conv2d-31 [-1, 512, 28, 28] 2,359,808
BatchNorm2d-32 [-1, 512, 28, 28] 1,024
ReLU-33 [-1, 512, 28, 28] 0
MaxPool2d-34 [-1, 512, 14, 14] 0
Conv2d-35 [-1, 512, 14, 14] 2,359,808
BatchNorm2d-36 [-1, 512, 14, 14] 1,024
ReLU-37 [-1, 512, 14, 14] 0
Conv2d-38 [-1, 512, 14, 14] 2,359,808
BatchNorm2d-39 [-1, 512, 14, 14] 1,024
ReLU-40 [-1, 512, 14, 14] 0
Conv2d-41 [-1, 512, 14, 14] 2,359,808
BatchNorm2d-42 [-1, 512, 14, 14] 1,024
ReLU-43 [-1, 512, 14, 14] 0
MaxPool2d-44 [-1, 512, 7, 7] 0
Dropout-45 [-1, 25088] 0
Linear-46 [-1, 4096] 102,764,544
ReLU-47 [-1, 4096] 0
Dropout-48 [-1, 4096] 0
Linear-49 [-1, 4096] 16,781,312
ReLU-50 [-1, 4096] 0
Linear-51 [-1, 3] 12,291
================================================================
Total params: 134,281,283
Trainable params: 134,281,283
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 322.10
Params size (MB): 512.24
Estimated Total Size (MB): 834.92
----------------------------------------------------------------
三、训练模型
1. 编写训练和测试函数
# 训练循环
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset) # 训练集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
train_loss, train_acc = 0, 0 # 初始化训练损失和正确率
for X, y in dataloader: # 获取图片及其标签
X, y = X.to(device), y.to(device)
# 计算预测误差
pred = model(X) # 网络输出
loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
# 反向传播
optimizer.zero_grad() # grad属性归零
loss.backward() # 反向传播
optimizer.step() # 每一步自动更新
# 记录acc与loss
train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
train_loss += loss.item()
train_acc /= size
train_loss /= num_batches
return train_acc, train_loss
## 测试函数和训练函数大致相同，但是由于不进行梯度下降对网络权重进行更新，所以不需要传入优化器
def test (dataloader, model, loss_fn):
size = len(dataloader.dataset) # 测试集的大小
num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)
test_loss, test_acc = 0, 0
# 当不进行训练时，停止梯度更新，节省计算内存消耗
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
# 计算loss
target_pred = model(imgs)
loss = loss_fn(target_pred, target)
test_loss += loss.item()
test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()
test_acc /= size
test_loss /= num_batches
return test_acc, test_loss
2. 正式训练
import copy
def tran_my_model(train_dl, test_dl, model, loss_fn, optimizer, epochs = 40, out_best_model = "best_model.pth"):
train_loss = []
train_acc = []
test_loss = []
test_acc = []
best_acc = 0 # 设置一个最佳准确率，作为最佳模型的判别指标
for epoch in range(epochs):
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
# 保存最佳模型到 best_model
if epoch_test_acc > best_acc:
best_acc = epoch_test_acc
best_model = copy.deepcopy(model)
train_acc.append(epoch_train_acc)
train_loss.append(epoch_train_loss)
test_acc.append(epoch_test_acc)
test_loss.append(epoch_test_loss)
# 获取当前的学习率
lr = optimizer.state_dict()['param_groups'][0]['lr']
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss,
epoch_test_acc*100, epoch_test_loss, lr))
torch.save(model.state_dict(), out_best_model)
return [best_model, train_loss, test_loss, train_acc, test_acc]
model = vgg16().to(device)
optimizer = torch.optim.Adam(model.parameters(), lr= 1e-4)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
adam_out = tran_my_model(train_dl, test_dl, model, loss_fn, optimizer, epochs = 40, out_best_model = "best_model.optim_Adam.pth")
adam_best_model, adam_train_loss, adam_test_loss, adam_train_acc, adam_test_acc = adam_out
Epoch: 1, Train_acc:48.3%, Train_loss:0.921, Test_acc:47.8%, Test_loss:0.912, Lr:1.00E-04
Epoch: 2, Train_acc:44.3%, Train_loss:0.910, Test_acc:45.2%, Test_loss:0.907, Lr:1.00E-04
Epoch: 3, Train_acc:45.3%, Train_loss:0.912, Test_acc:47.8%, Test_loss:0.892, Lr:1.00E-04
Epoch: 4, Train_acc:47.1%, Train_loss:0.897, Test_acc:47.8%, Test_loss:0.838, Lr:1.00E-04
Epoch: 5, Train_acc:78.8%, Train_loss:0.593, Test_acc:83.8%, Test_loss:0.473, Lr:1.00E-04
Epoch: 6, Train_acc:83.7%, Train_loss:0.449, Test_acc:83.8%, Test_loss:0.480, Lr:1.00E-04
Epoch: 7, Train_acc:86.6%, Train_loss:0.386, Test_acc:87.5%, Test_loss:0.418, Lr:1.00E-04
Epoch: 8, Train_acc:88.6%, Train_loss:0.322, Test_acc:90.7%, Test_loss:0.283, Lr:1.00E-04
Epoch: 9, Train_acc:89.3%, Train_loss:0.268, Test_acc:93.7%, Test_loss:0.203, Lr:1.00E-04
Epoch:10, Train_acc:91.8%, Train_loss:0.220, Test_acc:93.3%, Test_loss:0.164, Lr:1.00E-04
Epoch:11, Train_acc:93.3%, Train_loss:0.164, Test_acc:94.9%, Test_loss:0.133, Lr:1.00E-04
Epoch:12, Train_acc:95.1%, Train_loss:0.141, Test_acc:96.5%, Test_loss:0.100, Lr:1.00E-04
Epoch:13, Train_acc:95.4%, Train_loss:0.113, Test_acc:96.1%, Test_loss:0.100, Lr:1.00E-04
Epoch:14, Train_acc:96.0%, Train_loss:0.100, Test_acc:95.8%, Test_loss:0.092, Lr:1.00E-04
Epoch:15, Train_acc:96.4%, Train_loss:0.090, Test_acc:95.8%, Test_loss:0.130, Lr:1.00E-04
Epoch:16, Train_acc:97.0%, Train_loss:0.077, Test_acc:97.4%, Test_loss:0.073, Lr:1.00E-04
Epoch:17, Train_acc:96.8%, Train_loss:0.085, Test_acc:96.5%, Test_loss:0.114, Lr:1.00E-04
Epoch:18, Train_acc:96.7%, Train_loss:0.086, Test_acc:97.0%, Test_loss:0.084, Lr:1.00E-04
Epoch:19, Train_acc:97.4%, Train_loss:0.066, Test_acc:96.8%, Test_loss:0.094, Lr:1.00E-04
Epoch:20, Train_acc:97.3%, Train_loss:0.067, Test_acc:94.4%, Test_loss:0.146, Lr:1.00E-04
Epoch:21, Train_acc:96.3%, Train_loss:0.106, Test_acc:96.1%, Test_loss:0.109, Lr:1.00E-04
Epoch:22, Train_acc:97.2%, Train_loss:0.072, Test_acc:96.3%, Test_loss:0.120, Lr:1.00E-04
Epoch:23, Train_acc:98.3%, Train_loss:0.047, Test_acc:92.3%, Test_loss:0.301, Lr:1.00E-04
Epoch:24, Train_acc:97.0%, Train_loss:0.076, Test_acc:94.7%, Test_loss:0.225, Lr:1.00E-04
Epoch:25, Train_acc:98.2%, Train_loss:0.053, Test_acc:98.1%, Test_loss:0.081, Lr:1.00E-04
Epoch:26, Train_acc:97.3%, Train_loss:0.060, Test_acc:96.8%, Test_loss:0.072, Lr:1.00E-04
Epoch:27, Train_acc:98.3%, Train_loss:0.041, Test_acc:96.8%, Test_loss:0.092, Lr:1.00E-04
Epoch:28, Train_acc:98.4%, Train_loss:0.037, Test_acc:97.7%, Test_loss:0.076, Lr:1.00E-04
Epoch:29, Train_acc:98.1%, Train_loss:0.048, Test_acc:97.7%, Test_loss:0.096, Lr:1.00E-04
Epoch:30, Train_acc:99.2%, Train_loss:0.022, Test_acc:96.3%, Test_loss:0.149, Lr:1.00E-04
Epoch:31, Train_acc:99.1%, Train_loss:0.031, Test_acc:97.2%, Test_loss:0.085, Lr:1.00E-04
Epoch:32, Train_acc:99.3%, Train_loss:0.017, Test_acc:98.1%, Test_loss:0.078, Lr:1.00E-04
Epoch:33, Train_acc:99.0%, Train_loss:0.028, Test_acc:94.0%, Test_loss:0.183, Lr:1.00E-04
Epoch:34, Train_acc:98.8%, Train_loss:0.037, Test_acc:96.5%, Test_loss:0.134, Lr:1.00E-04
Epoch:35, Train_acc:99.4%, Train_loss:0.021, Test_acc:97.2%, Test_loss:0.096, Lr:1.00E-04
Epoch:36, Train_acc:99.4%, Train_loss:0.021, Test_acc:97.0%, Test_loss:0.126, Lr:1.00E-04
Epoch:37, Train_acc:99.2%, Train_loss:0.024, Test_acc:97.2%, Test_loss:0.106, Lr:1.00E-04
Epoch:38, Train_acc:99.4%, Train_loss:0.019, Test_acc:97.2%, Test_loss:0.124, Lr:1.00E-04
Epoch:39, Train_acc:99.2%, Train_loss:0.024, Test_acc:97.0%, Test_loss:0.117, Lr:1.00E-04
Epoch:40, Train_acc:99.9%, Train_loss:0.004, Test_acc:97.0%, Test_loss:0.145, Lr:1.00E-04
vgg16_model = VGG16Model(len(classeNames)).to(device) # 重置模型
optimizer = torch.optim.Adam(vgg16_model.parameters(), lr= 1e-4)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
vgg16_out = tran_my_model(train_dl, test_dl, vgg16_model, loss_fn, optimizer, epochs = 40, out_best_model = "VGG16.best_model.optim_Adam.pth")
vgg16_best_model, vgg16_train_loss, vgg16_test_loss, vgg16_train_acc, vgg16_test_acc = vgg16_out
Epoch: 1, Train_acc:81.8%, Train_loss:0.527, Test_acc:95.6%, Test_loss:0.122, Lr:1.00E-04
Epoch: 2, Train_acc:92.9%, Train_loss:0.202, Test_acc:92.1%, Test_loss:0.202, Lr:1.00E-04
Epoch: 3, Train_acc:95.4%, Train_loss:0.145, Test_acc:98.1%, Test_loss:0.068, Lr:1.00E-04
Epoch: 4, Train_acc:94.6%, Train_loss:0.153, Test_acc:97.9%, Test_loss:0.057, Lr:1.00E-04
Epoch: 5, Train_acc:94.8%, Train_loss:0.144, Test_acc:97.4%, Test_loss:0.060, Lr:1.00E-04
Epoch: 6, Train_acc:96.2%, Train_loss:0.108, Test_acc:96.8%, Test_loss:0.105, Lr:1.00E-04
Epoch: 7, Train_acc:97.2%, Train_loss:0.088, Test_acc:97.0%, Test_loss:0.083, Lr:1.00E-04
Epoch: 8, Train_acc:97.0%, Train_loss:0.088, Test_acc:98.6%, Test_loss:0.046, Lr:1.00E-04
Epoch: 9, Train_acc:97.7%, Train_loss:0.066, Test_acc:98.6%, Test_loss:0.026, Lr:1.00E-04
Epoch:10, Train_acc:97.2%, Train_loss:0.077, Test_acc:96.3%, Test_loss:0.113, Lr:1.00E-04
Epoch:11, Train_acc:95.6%, Train_loss:0.118, Test_acc:98.6%, Test_loss:0.044, Lr:1.00E-04
Epoch:12, Train_acc:98.5%, Train_loss:0.045, Test_acc:92.1%, Test_loss:0.182, Lr:1.00E-04
Epoch:13, Train_acc:98.1%, Train_loss:0.056, Test_acc:99.8%, Test_loss:0.013, Lr:1.00E-04
Epoch:14, Train_acc:98.5%, Train_loss:0.043, Test_acc:98.4%, Test_loss:0.038, Lr:1.00E-04
Epoch:15, Train_acc:98.5%, Train_loss:0.049, Test_acc:99.1%, Test_loss:0.025, Lr:1.00E-04
Epoch:16, Train_acc:99.0%, Train_loss:0.033, Test_acc:97.4%, Test_loss:0.077, Lr:1.00E-04
Epoch:17, Train_acc:96.9%, Train_loss:0.074, Test_acc:98.6%, Test_loss:0.046, Lr:1.00E-04
Epoch:18, Train_acc:98.7%, Train_loss:0.039, Test_acc:99.8%, Test_loss:0.008, Lr:1.00E-04
Epoch:19, Train_acc:98.8%, Train_loss:0.039, Test_acc:98.4%, Test_loss:0.028, Lr:1.00E-04
Epoch:20, Train_acc:99.2%, Train_loss:0.026, Test_acc:98.6%, Test_loss:0.047, Lr:1.00E-04
Epoch:21, Train_acc:98.5%, Train_loss:0.053, Test_acc:99.1%, Test_loss:0.039, Lr:1.00E-04
Epoch:22, Train_acc:98.7%, Train_loss:0.045, Test_acc:97.9%, Test_loss:0.058, Lr:1.00E-04
Epoch:23, Train_acc:99.4%, Train_loss:0.020, Test_acc:99.8%, Test_loss:0.009, Lr:1.00E-04
Epoch:24, Train_acc:98.5%, Train_loss:0.037, Test_acc:99.3%, Test_loss:0.016, Lr:1.00E-04
Epoch:25, Train_acc:98.8%, Train_loss:0.034, Test_acc:97.0%, Test_loss:0.095, Lr:1.00E-04
Epoch:26, Train_acc:99.2%, Train_loss:0.036, Test_acc:97.4%, Test_loss:0.113, Lr:1.00E-04
Epoch:27, Train_acc:99.1%, Train_loss:0.023, Test_acc:100.0%, Test_loss:0.001, Lr:1.00E-04
Epoch:28, Train_acc:98.4%, Train_loss:0.047, Test_acc:91.2%, Test_loss:0.191, Lr:1.00E-04
Epoch:29, Train_acc:99.5%, Train_loss:0.023, Test_acc:99.8%, Test_loss:0.004, Lr:1.00E-04
Epoch:30, Train_acc:99.9%, Train_loss:0.003, Test_acc:99.5%, Test_loss:0.010, Lr:1.00E-04
Epoch:31, Train_acc:98.0%, Train_loss:0.127, Test_acc:99.3%, Test_loss:0.040, Lr:1.00E-04
Epoch:32, Train_acc:99.6%, Train_loss:0.015, Test_acc:99.3%, Test_loss:0.027, Lr:1.00E-04
Epoch:33, Train_acc:99.0%, Train_loss:0.039, Test_acc:99.1%, Test_loss:0.028, Lr:1.00E-04
Epoch:34, Train_acc:99.4%, Train_loss:0.017, Test_acc:99.3%, Test_loss:0.017, Lr:1.00E-04
Epoch:35, Train_acc:99.7%, Train_loss:0.007, Test_acc:99.5%, Test_loss:0.020, Lr:1.00E-04
Epoch:36, Train_acc:99.4%, Train_loss:0.038, Test_acc:90.5%, Test_loss:0.343, Lr:1.00E-04
Epoch:37, Train_acc:98.8%, Train_loss:0.045, Test_acc:100.0%, Test_loss:0.003, Lr:1.00E-04
Epoch:38, Train_acc:99.2%, Train_loss:0.023, Test_acc:99.8%, Test_loss:0.004, Lr:1.00E-04
Epoch:39, Train_acc:99.7%, Train_loss:0.018, Test_acc:100.0%, Test_loss:0.002, Lr:1.00E-04
Epoch:40, Train_acc:99.6%, Train_loss:0.010, Test_acc:100.0%, Test_loss:0.003, Lr:1.00E-04
model = vgg16().to(device) # 重置模型
## 测试如果优化器变成SGD的结果
learn_rate = 1e-4
optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
sgd_out = tran_my_model(train_dl, test_dl, model, loss_fn, optimizer, epochs = 40, out_best_model = "best_model.optim_SGD.pth")
sgd_best_model, sgd_train_loss, sgd_test_loss, sgd_train_acc, sgd_test_acc = sgd_out
Epoch: 1, Train_acc:15.6%, Train_loss:1.100, Test_acc:45.2%, Test_loss:1.099, Lr:1.00E-04
Epoch: 2, Train_acc:46.8%, Train_loss:1.099, Test_acc:45.2%, Test_loss:1.099, Lr:1.00E-04
Epoch: 3, Train_acc:46.8%, Train_loss:1.098, Test_acc:45.2%, Test_loss:1.098, Lr:1.00E-04
Epoch: 4, Train_acc:46.8%, Train_loss:1.097, Test_acc:45.2%, Test_loss:1.097, Lr:1.00E-04
Epoch: 5, Train_acc:46.8%, Train_loss:1.097, Test_acc:45.2%, Test_loss:1.096, Lr:1.00E-04
Epoch: 6, Train_acc:46.8%, Train_loss:1.096, Test_acc:45.2%, Test_loss:1.095, Lr:1.00E-04
Epoch: 7, Train_acc:46.8%, Train_loss:1.095, Test_acc:45.2%, Test_loss:1.095, Lr:1.00E-04
Epoch: 8, Train_acc:46.8%, Train_loss:1.094, Test_acc:45.2%, Test_loss:1.094, Lr:1.00E-04
Epoch: 9, Train_acc:46.8%, Train_loss:1.093, Test_acc:45.2%, Test_loss:1.093, Lr:1.00E-04
Epoch:10, Train_acc:46.8%, Train_loss:1.093, Test_acc:45.2%, Test_loss:1.092, Lr:1.00E-04
Epoch:11, Train_acc:46.8%, Train_loss:1.092, Test_acc:45.2%, Test_loss:1.092, Lr:1.00E-04
Epoch:12, Train_acc:46.8%, Train_loss:1.091, Test_acc:45.2%, Test_loss:1.091, Lr:1.00E-04
Epoch:13, Train_acc:46.8%, Train_loss:1.090, Test_acc:45.2%, Test_loss:1.090, Lr:1.00E-04
Epoch:14, Train_acc:46.8%, Train_loss:1.090, Test_acc:45.2%, Test_loss:1.089, Lr:1.00E-04
Epoch:15, Train_acc:46.8%, Train_loss:1.089, Test_acc:45.2%, Test_loss:1.089, Lr:1.00E-04
Epoch:16, Train_acc:46.8%, Train_loss:1.088, Test_acc:45.2%, Test_loss:1.088, Lr:1.00E-04
Epoch:17, Train_acc:46.8%, Train_loss:1.087, Test_acc:45.2%, Test_loss:1.087, Lr:1.00E-04
Epoch:18, Train_acc:46.8%, Train_loss:1.087, Test_acc:45.2%, Test_loss:1.087, Lr:1.00E-04
Epoch:19, Train_acc:46.8%, Train_loss:1.086, Test_acc:45.2%, Test_loss:1.086, Lr:1.00E-04
Epoch:20, Train_acc:46.8%, Train_loss:1.085, Test_acc:45.2%, Test_loss:1.085, Lr:1.00E-04
Epoch:21, Train_acc:46.8%, Train_loss:1.085, Test_acc:45.2%, Test_loss:1.084, Lr:1.00E-04
Epoch:22, Train_acc:46.8%, Train_loss:1.084, Test_acc:45.2%, Test_loss:1.084, Lr:1.00E-04
Epoch:23, Train_acc:46.8%, Train_loss:1.083, Test_acc:45.2%, Test_loss:1.083, Lr:1.00E-04
Epoch:24, Train_acc:46.8%, Train_loss:1.082, Test_acc:45.2%, Test_loss:1.082, Lr:1.00E-04
Epoch:25, Train_acc:46.8%, Train_loss:1.082, Test_acc:45.2%, Test_loss:1.082, Lr:1.00E-04
Epoch:26, Train_acc:46.8%, Train_loss:1.081, Test_acc:45.2%, Test_loss:1.081, Lr:1.00E-04
Epoch:27, Train_acc:46.8%, Train_loss:1.080, Test_acc:45.2%, Test_loss:1.080, Lr:1.00E-04
Epoch:28, Train_acc:46.8%, Train_loss:1.080, Test_acc:45.2%, Test_loss:1.079, Lr:1.00E-04
Epoch:29, Train_acc:46.8%, Train_loss:1.079, Test_acc:45.2%, Test_loss:1.079, Lr:1.00E-04
Epoch:30, Train_acc:46.8%, Train_loss:1.078, Test_acc:45.2%, Test_loss:1.078, Lr:1.00E-04
Epoch:31, Train_acc:46.8%, Train_loss:1.078, Test_acc:45.2%, Test_loss:1.077, Lr:1.00E-04
Epoch:32, Train_acc:46.8%, Train_loss:1.077, Test_acc:45.2%, Test_loss:1.077, Lr:1.00E-04
Epoch:33, Train_acc:46.8%, Train_loss:1.076, Test_acc:45.2%, Test_loss:1.077, Lr:1.00E-04
Epoch:34, Train_acc:46.8%, Train_loss:1.075, Test_acc:45.2%, Test_loss:1.075, Lr:1.00E-04
Epoch:35, Train_acc:46.8%, Train_loss:1.075, Test_acc:45.2%, Test_loss:1.074, Lr:1.00E-04
Epoch:36, Train_acc:46.8%, Train_loss:1.074, Test_acc:45.2%, Test_loss:1.074, Lr:1.00E-04
Epoch:37, Train_acc:46.8%, Train_loss:1.074, Test_acc:45.2%, Test_loss:1.073, Lr:1.00E-04
Epoch:38, Train_acc:46.8%, Train_loss:1.073, Test_acc:45.2%, Test_loss:1.072, Lr:1.00E-04
Epoch:39, Train_acc:46.8%, Train_loss:1.072, Test_acc:45.2%, Test_loss:1.072, Lr:1.00E-04
Epoch:40, Train_acc:46.8%, Train_loss:1.072, Test_acc:45.2%, Test_loss:1.071, Lr:1.00E-04
vgg16_model = VGG16Model(len(classeNames)).to(device) # 重置模型
optimizer = torch.optim.SGD(vgg16_model.parameters(), lr= 1e-4)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
vgg16_out2 = tran_my_model(train_dl, test_dl, vgg16_model, loss_fn, optimizer, epochs = 40, out_best_model = "VGG16.best_model.optim_SGD.pth")
vgg16_best_model2, vgg16_train_loss2, vgg16_test_loss2, vgg16_train_acc2, vgg16_test_acc2 = vgg16_out2
Epoch: 1, Train_acc:52.3%, Train_loss:0.959, Test_acc:63.3%, Test_loss:0.926, Lr:1.00E-04
Epoch: 2, Train_acc:61.2%, Train_loss:0.860, Test_acc:61.9%, Test_loss:0.824, Lr:1.00E-04
Epoch: 3, Train_acc:65.4%, Train_loss:0.816, Test_acc:73.5%, Test_loss:0.770, Lr:1.00E-04
Epoch: 4, Train_acc:73.4%, Train_loss:0.771, Test_acc:77.3%, Test_loss:0.727, Lr:1.00E-04
Epoch: 5, Train_acc:76.7%, Train_loss:0.734, Test_acc:79.6%, Test_loss:0.692, Lr:1.00E-04
Epoch: 6, Train_acc:80.7%, Train_loss:0.703, Test_acc:82.1%, Test_loss:0.644, Lr:1.00E-04
Epoch: 7, Train_acc:82.1%, Train_loss:0.667, Test_acc:84.0%, Test_loss:0.620, Lr:1.00E-04
Epoch: 8, Train_acc:83.1%, Train_loss:0.641, Test_acc:84.9%, Test_loss:0.580, Lr:1.00E-04
Epoch: 9, Train_acc:85.2%, Train_loss:0.607, Test_acc:86.1%, Test_loss:0.543, Lr:1.00E-04
Epoch:10, Train_acc:85.9%, Train_loss:0.579, Test_acc:86.8%, Test_loss:0.522, Lr:1.00E-04
Epoch:11, Train_acc:85.8%, Train_loss:0.553, Test_acc:86.8%, Test_loss:0.492, Lr:1.00E-04
Epoch:12, Train_acc:86.2%, Train_loss:0.535, Test_acc:86.8%, Test_loss:0.469, Lr:1.00E-04
Epoch:13, Train_acc:86.7%, Train_loss:0.511, Test_acc:87.2%, Test_loss:0.453, Lr:1.00E-04
Epoch:14, Train_acc:85.8%, Train_loss:0.503, Test_acc:87.0%, Test_loss:0.449, Lr:1.00E-04
Epoch:15, Train_acc:87.2%, Train_loss:0.481, Test_acc:86.5%, Test_loss:0.423, Lr:1.00E-04
Epoch:16, Train_acc:86.5%, Train_loss:0.465, Test_acc:86.8%, Test_loss:0.419, Lr:1.00E-04
Epoch:17, Train_acc:86.9%, Train_loss:0.450, Test_acc:87.2%, Test_loss:0.397, Lr:1.00E-04
Epoch:18, Train_acc:87.4%, Train_loss:0.436, Test_acc:87.0%, Test_loss:0.400, Lr:1.00E-04
Epoch:19, Train_acc:86.8%, Train_loss:0.424, Test_acc:86.5%, Test_loss:0.390, Lr:1.00E-04
Epoch:20, Train_acc:87.7%, Train_loss:0.413, Test_acc:86.5%, Test_loss:0.365, Lr:1.00E-04
Epoch:21, Train_acc:87.4%, Train_loss:0.409, Test_acc:86.5%, Test_loss:0.360, Lr:1.00E-04
Epoch:22, Train_acc:87.7%, Train_loss:0.400, Test_acc:86.8%, Test_loss:0.362, Lr:1.00E-04
Epoch:23, Train_acc:88.0%, Train_loss:0.385, Test_acc:87.0%, Test_loss:0.347, Lr:1.00E-04
Epoch:24, Train_acc:88.0%, Train_loss:0.382, Test_acc:87.2%, Test_loss:0.348, Lr:1.00E-04
Epoch:25, Train_acc:88.5%, Train_loss:0.374, Test_acc:87.2%, Test_loss:0.341, Lr:1.00E-04
Epoch:26, Train_acc:88.4%, Train_loss:0.363, Test_acc:87.0%, Test_loss:0.340, Lr:1.00E-04
Epoch:27, Train_acc:87.5%, Train_loss:0.367, Test_acc:87.0%, Test_loss:0.326, Lr:1.00E-04
Epoch:28, Train_acc:88.7%, Train_loss:0.353, Test_acc:87.0%, Test_loss:0.320, Lr:1.00E-04
Epoch:29, Train_acc:88.1%, Train_loss:0.353, Test_acc:88.6%, Test_loss:0.297, Lr:1.00E-04
Epoch:30, Train_acc:88.9%, Train_loss:0.337, Test_acc:88.4%, Test_loss:0.306, Lr:1.00E-04
Epoch:31, Train_acc:88.7%, Train_loss:0.339, Test_acc:87.7%, Test_loss:0.299, Lr:1.00E-04
Epoch:32, Train_acc:88.7%, Train_loss:0.331, Test_acc:89.6%, Test_loss:0.280, Lr:1.00E-04
Epoch:33, Train_acc:88.5%, Train_loss:0.318, Test_acc:89.6%, Test_loss:0.272, Lr:1.00E-04
Epoch:34, Train_acc:88.8%, Train_loss:0.321, Test_acc:89.6%, Test_loss:0.265, Lr:1.00E-04
Epoch:35, Train_acc:89.1%, Train_loss:0.315, Test_acc:89.6%, Test_loss:0.268, Lr:1.00E-04
Epoch:36, Train_acc:89.1%, Train_loss:0.313, Test_acc:89.3%, Test_loss:0.256, Lr:1.00E-04
Epoch:37, Train_acc:89.5%, Train_loss:0.298, Test_acc:90.0%, Test_loss:0.251, Lr:1.00E-04
Epoch:38, Train_acc:88.9%, Train_loss:0.313, Test_acc:89.6%, Test_loss:0.252, Lr:1.00E-04
Epoch:39, Train_acc:89.9%, Train_loss:0.287, Test_acc:89.6%, Test_loss:0.239, Lr:1.00E-04
Epoch:40, Train_acc:89.6%, Train_loss:0.285, Test_acc:89.8%, Test_loss:0.230, Lr:1.00E-04
四、结果可视化
1. Loss与Accuracy图
import matplotlib.pyplot as plt
import warnings
def plot_acc_loss(epoch_acc, epoch_loss):
warnings.filterwarnings("ignore") #忽略警告信息
#plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
plt.rcParams['figure.dpi'] = 100 #分辨率
train_acc, test_acc = epoch_acc
train_loss, test_loss = epoch_loss
epochs_range = range(len(train_acc))
plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)
plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()
plot_acc_loss([adam_train_acc, adam_test_acc], [adam_train_loss, adam_test_loss])
plot_acc_loss([vgg16_train_acc, vgg16_test_acc], [vgg16_train_loss, vgg16_test_loss])
plot_acc_loss([sgd_train_acc, sgd_test_acc], [sgd_train_loss, sgd_test_loss])
- 当设为SDG为优化器的时候，VGG16模型的误差不更新，目前不清楚具体原因。
2. 指定图片进行预测
from PIL import Image
classes = list(total_data.class_to_idx)
def predict_one_image(image_path, model, transform, classes):
test_img = Image.open(image_path).convert('RGB')
plt.imshow(test_img) # 展示预测的图片
test_img = transform(test_img)
img = test_img.to(device).unsqueeze(0)
model.eval()
output = model(img)
_,pred = torch.max(output,1)
pred_class = classes[pred]
print(f'预测结果是：{pred_class}')
3. 模型评估
best_model = vgg16_best_model
best_model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)
epoch_test_acc, epoch_test_loss
(1.0, 0.001758532521957282)
五、拔高训练
1. 验证集准确率达到100%
- 当在全连接层加入nn.Dropout(0.5)的时候，可以使验证集的准确率达到100%
2. 动手画出VGG-16算法框架图
- 找到一些自动画算法框架图的资源：github.com/ashishpatel…
3. 轻量化模型
- 尝试减少全连接层的参数个数
class vgg16_reduced(nn.Module):
def __init__(self):
super(vgg16_reduced, self).__init__()
# 卷积块1
self.block1 = nn.Sequential(
nn.Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块2
self.block2 = nn.Sequential(
nn.Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块3
self.block3 = nn.Sequential(
nn.Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块4
self.block4 = nn.Sequential(
nn.Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 卷积块5
self.block5 = nn.Sequential(
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
# 全连接网络层，用于分类
self.classifier = nn.Sequential(
nn.Dropout(0.5),
nn.Linear(in_features=512*7*7, out_features=1024),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(in_features=1024, out_features=128),
nn.ReLU(),
nn.Linear(in_features=128, out_features=3)
)
def forward(self, x):
x = self.block1(x)
x = self.block2(x)
x = self.block3(x)
x = self.block4(x)
x = self.block5(x)
x = torch.flatten(x, start_dim=1)
x = self.classifier(x)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
model_reduced = vgg16_reduced().to(device)
model_reduced
Using cuda device
vgg16_reduced(
(block1): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block2): Sequential(
(0): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block3): Sequential(
(0): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block4): Sequential(
(0): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(block5): Sequential(
(0): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(5): ReLU()
(6): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
)
(classifier): Sequential(
(0): Dropout(p=0.5, inplace=False)
(1): Linear(in_features=25088, out_features=1024, bias=True)
(2): ReLU()
(3): Dropout(p=0.5, inplace=False)
(4): Linear(in_features=1024, out_features=128, bias=True)
(5): ReLU()
(6): Linear(in_features=128, out_features=3, bias=True)
)
)
import torchsummary as summary
summary.summary(model_reduced, (3, 224, 224))
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 224, 224] 1,792
ReLU-2 [-1, 64, 224, 224] 0
Conv2d-3 [-1, 64, 224, 224] 36,928
ReLU-4 [-1, 64, 224, 224] 0
MaxPool2d-5 [-1, 64, 112, 112] 0
Conv2d-6 [-1, 128, 112, 112] 73,856
ReLU-7 [-1, 128, 112, 112] 0
Conv2d-8 [-1, 128, 112, 112] 147,584
ReLU-9 [-1, 128, 112, 112] 0
MaxPool2d-10 [-1, 128, 56, 56] 0
Conv2d-11 [-1, 256, 56, 56] 295,168
ReLU-12 [-1, 256, 56, 56] 0
Conv2d-13 [-1, 256, 56, 56] 590,080
ReLU-14 [-1, 256, 56, 56] 0
Conv2d-15 [-1, 256, 56, 56] 590,080
ReLU-16 [-1, 256, 56, 56] 0
MaxPool2d-17 [-1, 256, 28, 28] 0
Conv2d-18 [-1, 512, 28, 28] 1,180,160
ReLU-19 [-1, 512, 28, 28] 0
Conv2d-20 [-1, 512, 28, 28] 2,359,808
ReLU-21 [-1, 512, 28, 28] 0
Conv2d-22 [-1, 512, 28, 28] 2,359,808
ReLU-23 [-1, 512, 28, 28] 0
MaxPool2d-24 [-1, 512, 14, 14] 0
Conv2d-25 [-1, 512, 14, 14] 2,359,808
ReLU-26 [-1, 512, 14, 14] 0
Conv2d-27 [-1, 512, 14, 14] 2,359,808
ReLU-28 [-1, 512, 14, 14] 0
Conv2d-29 [-1, 512, 14, 14] 2,359,808
ReLU-30 [-1, 512, 14, 14] 0
MaxPool2d-31 [-1, 512, 7, 7] 0
Dropout-32 [-1, 25088] 0
Linear-33 [-1, 1024] 25,691,136
ReLU-34 [-1, 1024] 0
Dropout-35 [-1, 1024] 0
Linear-36 [-1, 128] 131,200
ReLU-37 [-1, 128] 0
Linear-38 [-1, 3] 387
================================================================
Total params: 40,537,411
Trainable params: 40,537,411
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 218.61
Params size (MB): 154.64
Estimated Total Size (MB): 373.82
----------------------------------------------------------------
learn_rate = 1e-4
optimizer = torch.optim.Adam(model_reduced.parameters(), lr=learn_rate)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
reduced_out = tran_my_model(train_dl, test_dl, model_reduced, loss_fn, optimizer, epochs = 40, out_best_model = "best_model.optim_SGD.pth")
reduced_best_model, reduced_train_loss, reduced_test_loss, reduced_train_acc, reduced_test_acc = reduced_out
Epoch: 1, Train_acc:45.3%, Train_loss:0.950, Test_acc:45.2%, Test_loss:0.895, Lr:1.00E-04
Epoch: 2, Train_acc:58.9%, Train_loss:0.746, Test_acc:64.5%, Test_loss:0.774, Lr:1.00E-04
Epoch: 3, Train_acc:79.1%, Train_loss:0.580, Test_acc:82.1%, Test_loss:0.483, Lr:1.00E-04
Epoch: 4, Train_acc:83.9%, Train_loss:0.453, Test_acc:84.2%, Test_loss:0.461, Lr:1.00E-04
Epoch: 5, Train_acc:85.1%, Train_loss:0.418, Test_acc:84.5%, Test_loss:0.444, Lr:1.00E-04
Epoch: 6, Train_acc:85.2%, Train_loss:0.408, Test_acc:86.5%, Test_loss:0.350, Lr:1.00E-04
Epoch: 7, Train_acc:89.1%, Train_loss:0.285, Test_acc:90.0%, Test_loss:0.258, Lr:1.00E-04
Epoch: 8, Train_acc:88.0%, Train_loss:0.317, Test_acc:90.7%, Test_loss:0.212, Lr:1.00E-04
Epoch: 9, Train_acc:89.2%, Train_loss:0.295, Test_acc:90.3%, Test_loss:0.232, Lr:1.00E-04
Epoch:10, Train_acc:91.9%, Train_loss:0.200, Test_acc:90.5%, Test_loss:0.297, Lr:1.00E-04
Epoch:11, Train_acc:92.6%, Train_loss:0.184, Test_acc:89.6%, Test_loss:0.423, Lr:1.00E-04
Epoch:12, Train_acc:92.3%, Train_loss:0.192, Test_acc:95.6%, Test_loss:0.108, Lr:1.00E-04
Epoch:13, Train_acc:93.6%, Train_loss:0.160, Test_acc:96.3%, Test_loss:0.110, Lr:1.00E-04
Epoch:14, Train_acc:95.6%, Train_loss:0.118, Test_acc:97.2%, Test_loss:0.091, Lr:1.00E-04
Epoch:15, Train_acc:96.8%, Train_loss:0.088, Test_acc:94.9%, Test_loss:0.151, Lr:1.00E-04
Epoch:16, Train_acc:96.0%, Train_loss:0.108, Test_acc:95.6%, Test_loss:0.163, Lr:1.00E-04
Epoch:17, Train_acc:97.1%, Train_loss:0.089, Test_acc:98.4%, Test_loss:0.060, Lr:1.00E-04
Epoch:18, Train_acc:93.4%, Train_loss:0.194, Test_acc:95.1%, Test_loss:0.138, Lr:1.00E-04
Epoch:19, Train_acc:96.5%, Train_loss:0.102, Test_acc:97.4%, Test_loss:0.087, Lr:1.00E-04
Epoch:20, Train_acc:96.5%, Train_loss:0.098, Test_acc:97.7%, Test_loss:0.077, Lr:1.00E-04
Epoch:21, Train_acc:97.9%, Train_loss:0.066, Test_acc:98.8%, Test_loss:0.037, Lr:1.00E-04
Epoch:22, Train_acc:97.7%, Train_loss:0.058, Test_acc:96.8%, Test_loss:0.091, Lr:1.00E-04
Epoch:23, Train_acc:98.4%, Train_loss:0.051, Test_acc:97.0%, Test_loss:0.083, Lr:1.00E-04
Epoch:24, Train_acc:98.5%, Train_loss:0.048, Test_acc:98.8%, Test_loss:0.029, Lr:1.00E-04
Epoch:25, Train_acc:98.8%, Train_loss:0.032, Test_acc:98.4%, Test_loss:0.057, Lr:1.00E-04
Epoch:26, Train_acc:98.8%, Train_loss:0.036, Test_acc:97.0%, Test_loss:0.081, Lr:1.00E-04
Epoch:27, Train_acc:98.7%, Train_loss:0.043, Test_acc:98.6%, Test_loss:0.061, Lr:1.00E-04
Epoch:28, Train_acc:99.4%, Train_loss:0.023, Test_acc:98.8%, Test_loss:0.062, Lr:1.00E-04
Epoch:29, Train_acc:99.0%, Train_loss:0.038, Test_acc:95.8%, Test_loss:0.125, Lr:1.00E-04
Epoch:30, Train_acc:99.1%, Train_loss:0.029, Test_acc:98.8%, Test_loss:0.035, Lr:1.00E-04
Epoch:31, Train_acc:98.8%, Train_loss:0.027, Test_acc:98.6%, Test_loss:0.049, Lr:1.00E-04
Epoch:32, Train_acc:99.1%, Train_loss:0.027, Test_acc:99.3%, Test_loss:0.043, Lr:1.00E-04
Epoch:33, Train_acc:99.0%, Train_loss:0.038, Test_acc:97.2%, Test_loss:0.101, Lr:1.00E-04
Epoch:34, Train_acc:99.0%, Train_loss:0.031, Test_acc:99.1%, Test_loss:0.060, Lr:1.00E-04
Epoch:35, Train_acc:99.4%, Train_loss:0.018, Test_acc:98.6%, Test_loss:0.036, Lr:1.00E-04
Epoch:36, Train_acc:98.3%, Train_loss:0.065, Test_acc:98.4%, Test_loss:0.036, Lr:1.00E-04
Epoch:37, Train_acc:99.4%, Train_loss:0.019, Test_acc:98.8%, Test_loss:0.044, Lr:1.00E-04
Epoch:38, Train_acc:99.4%, Train_loss:0.014, Test_acc:98.8%, Test_loss:0.049, Lr:1.00E-04
Epoch:39, Train_acc:99.6%, Train_loss:0.017, Test_acc:98.1%, Test_loss:0.105, Lr:1.00E-04
Epoch:40, Train_acc:98.5%, Train_loss:0.058, Test_acc:99.3%, Test_loss:0.024, Lr:1.00E-04
plot_acc_loss([reduced_train_acc, reduced_test_acc], [reduced_train_loss, reduced_test_loss])
六、总结
- 通过降低模型内全连接层的参数数目，总参数个数为40,537,411，测试集准确性还可以达到99.3%。
