计算机视觉实验 - 卷积神经网络 PyTorch试水

edwin99
edwin99
2024-02-04 01:23
31 阅读
0 评论
文章封面
目录
正在加载目录...

卷积神经网络：找图像中特定物体

 

准备工作：

import torch

import torch.nn as nn

import torchvision

import matplotlib.pyplot as plt

from torchinfo import summary

import numpy as np

 

from pytorchcv import load_mnist, train, plot_results, plot_convolution, display_datasetload_mnist(batch_size=128)

卷积过滤器：在每个像素移动的小窗口，计算邻接像素的平均W，提取物体pattern；由权系数矩阵定义；

 

MNIST数字用到的两种过滤器（水平，垂直）：

plot_convolution(torch.tensor([[-1.,0.,1.],[-1.,0.,1.],[-1.,0.,1.]]),'Vertical edge filter')plot_convolution(torch.tensor([[-1.,-1.,-1.],[0.,0.,0.],[1.,1.,1.]]),'Horizontal edge filter')

 

水平边缘过滤器：每处理一个均匀像素域就置零

碰到垂直边缘就置高值，水平边缘被平均掉；

垂直边缘过滤器：

反过来增强水平线条，垂直边缘被平均

传统CV vs 深度学习：

  1. 传统要手动设多滤波器，再特征提取，再分类器

  2. 深度学习的网络自动学习最优滤波器，再端对端分类

 

卷积层：nn.Conv2d；参数有3ge1（in_channels=1灰度输入通道数，out_channels=9用9个滤波器，kernel_size=5x5滑动窗口）

输出输入尺寸：28x28卷积后输出张量9x24x24（因为28-5+1=24）

网络结构：卷积层展平后9x24x24=5184维向量，变成线性层输出10类，中间用ReLU激活

总结：滤波器#决定特征提取，卷积核大小影响（不是决定）输出尺寸

代码：

class OneConv(nn.Module):

def __init__(self):

super(OneConv, self).__init__()

self.conv = nn.Conv2d(in_channels=1,out_channels=9,kernel_size=(5,5))

self.flatten = nn.Flatten()

self.fc = nn.Linear(5184,10)

 

def forward(self, x):

x = nn.functional.relu(self.conv(x))

x = self.flatten(x)

x = nn.functional.log_softmax(self.fc(x),dim=1)

return x

 

net = OneConv()

 

summary(net,input_size=(1,1,28,28))

输出：

========================================================================================== Layer (type:depth-idx) Output Shape Param # ========================================================================================== ├─Conv2d: 1-1 [1, 9, 24, 24] 234 ├─Flatten: 1-2 [1, 5184] -- ├─Linear: 1-3 [1, 10] 51,850 ========================================================================================== Total params: 52,084 Trainable params: 52,084 Non-trainable params: 0 Total mult-adds (M): 0.18 ========================================================================================== Input size (MB): 0.00 Forward/backward pass size (MB): 0.04 Params size (MB): 0.21 Estimated Total Size (MB): 0.25 ==========================================================================================

 

结果：网络含50k训练参数，相比80k全连接多层神经网络，可在小数据集存储更加结果

绘图：

hist = train(net,train_loader,test_loader,epochs=5)

plot_results(hist)

结果：

Epoch 0, Train acc=0.947, Val acc=0.969, Train loss=0.001, Val loss=0.001

Epoch 1, Train acc=0.979, Val acc=0.975, Train loss=0.001, Val loss=0.001

Epoch 2, Train acc=0.985, Val acc=0.977, Train loss=0.000, Val loss=0.001

Epoch 3, Train acc=0.988, Val acc=0.975, Train loss=0.000, Val loss=0.001

Epoch 4, Train acc=0.988, Val acc=0.976, Train loss=0.000, Val loss=0.001

可视化训练卷积层的权：

fig,ax = plt.subplots(1,9)with torch.no_grad():

p = next(net.conv.parameters())

for i,x in enumerate(p):

ax[i].imshow(x.detach().cpu()[0,...])

ax[i].axis('off')

卷积层：检测图像pattern；过程：初层检测简单pattern（线条）-后续层结合pattern+检测复杂特征（形状）-最终层对物体分类

 

池化层：减少图像空间尺寸+降低计算量+防止过拟合，2种

  1. 平均池化：滑动窗口2x2pix算平均值

  2. 最大池化：用最大值替换窗口，检测pattern

CNN架构：层次（在多卷积层间插入池化层），特征图（空间尺寸逐减，滤波器数逐增），输出（用全连接层+激活函数来分类）

 

金字塔架构：空间尺寸逐减，滤波器数逐增，用来高效提取分类特征

分析：输入1x28x28图像 - 第一卷积层 - 第二卷积层 - 输出（全输出层，LogSOftmax用来分10类）

（每个卷积层都是2x2窗口，10个滤波器，R额LU激活函数，最大池化）

代码:

class MultiLayerCNN(nn.Module):

def __init__(self):

super(MultiLayerCNN, self).__init__()

self.conv1 = nn.Conv2d(1, 10, 5)

self.pool = nn.MaxPool2d(2, 2)

self.conv2 = nn.Conv2d(10, 20, 5)

self.fc = nn.Linear(320,10)

 

def forward(self, x):

x = self.pool(nn.functional.relu(self.conv1(x)))

x = self.pool(nn.functional.relu(self.conv2(x)))

x = x.view(-1, 320)

x = nn.functional.log_softmax(self.fc(x),dim=1)

return x

 

net = MultiLayerCNN()summary(net,input_size=(1,1,28,28))

输出：

========================================================================================== Layer (type:depth-idx) Output Shape Param # ========================================================================================== ├─Conv2d: 1-1 [1, 10, 24, 24] 260 ├─MaxPool2d: 1-2 [1, 10, 12, 12] -- ├─Conv2d: 1-3 [1, 20, 8, 8] 5,020 ├─MaxPool2d: 1-4 [1, 20, 4, 4] -- ├─Linear: 1-5 [1, 10] 3,210 ========================================================================================== Total params: 8,490 Trainable params: 8,490 Non-trainable params: 0 Total mult-adds (M): 0.47 ========================================================================================== Input size (MB): 0.00 Forward/backward pass size (MB): 0.06 Params size (MB): 0.03 Estimated Total Size (MB): 0.09 ==========================================================================================

 

Flatten：在forward用view展平张量，不用单独层

Pooling：无参数可复用

参数：少，~8.5k，防止、过拟合

 

代码：

hist = train(net,train_loader,test_loader,epochs=5)

 

结果：一层精准度更高更快，只用1-2个epoch

Epoch 0, Train acc=0.952, Val acc=0.977, Train loss=0.001, Val loss=0.001

Epoch 1, Train acc=0.982, Val acc=0.983, Train loss=0.000, Val loss=0.000

Epoch 2, Train acc=0.986, Val acc=0.983, Train loss=0.000, Val loss=0.000

Epoch 3, Train acc=0.986, Val acc=0.978, Train loss=0.000, Val loss=0.001

Epoch 4, Train acc=0.987, Val acc=0.981, Train loss=0.000, Val loss=0.000

 

 

用CIFAR-10数据集：60000张图（32x32，10类：飞机-车-鸟-猫-鹿-狗-蛙-马-船-卡车，比手写数字复杂），常用架构：LeNet

transform = torchvision.transforms.Compose(

[torchvision.transforms.ToTensor(),

torchvision.transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

 

trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform)trainloader = torch.utils.data.DataLoader(trainset, batch_size=14, shuffle=True)testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform)testloader = torch.utils.data.DataLoader(testset, batch_size=14, shuffle=False)classes = ('plane', 'car', 'bird', 'cat',

'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

display_dataset(trainset,classes=classes)

 

LeNet架构：输入RGB，输出层不用long_softmax，直返全连接层输出，损失函数用CrossEntropyLoss优化模型

 

class LeNet(nn.Module):

def __init__(self):

super(LeNet, self).__init__()

self.conv1 = nn.Conv2d(3, 6, 5)

self.pool = nn.MaxPool2d(2)

self.conv2 = nn.Conv2d(6, 16, 5)

self.conv3 = nn.Conv2d(16,120,5)

self.flat = nn.Flatten()

self.fc1 = nn.Linear(120,64)

self.fc2 = nn.Linear(64,10)

 

def forward(self, x):

x = self.pool(nn.functional.relu(self.conv1(x)))

x = self.pool(nn.functional.relu(self.conv2(x)))

x = nn.functional.relu(self.conv3(x))

x = self.flat(x)

x = nn.functional.relu(self.fc1(x))

x = self.fc2(x)

return x

 

net = LeNet()

 

summary(net,input_size=(1,3,32,32))

opt = torch.optim.SGD(net.parameters(),lr=0.001,momentum=0.9)

hist = train(net, trainloader, testloader, epochs=3, optimizer=opt, loss_fn=nn.CrossEntropyLoss())

输出：3个epoch，复杂模型50%已算不错

Epoch 0, Train acc=0.261, Val acc=0.388, Train loss=0.143, Val loss=0.121 Epoch 1, Train acc=0.437, Val acc=0.491, Train loss=0.110, Val loss=0.101 Epoch 2, Train acc=0.508, Val acc=0.522, Train loss=0.097, Val loss=0.094

 

扩展阅读：

  1. 识别宠物脸部实验

  2. 卷积神经网络PyTorch试水

  3. OPenCV实验

  4. 用光流实现手掌运动检测

 

 

 

 

机器学习 神经网络 PyTorch CNN

评论区 (0)

请登录后参与评论

暂无评论，抢沙发吧！