系统学习CV-CNN

参考

https://zh-v2.d2l.ai/index.html

CNN卷积神经网络

概念

性质

空间不变性
平移不变性：不管目标出现在图像汇总的哪个位置，神经网络的前几层应该对相同的图像区域具有相似的响应，即平移不变性
局部性：神经网络的前面几层应该只探索输入图像中的局部区域，而不过度关注图像中相隔较远区域的关系，即局部性。最终聚合这些局部特征，以在整个图像级别进行预测

互相关

约定俗成为卷积运算

感受野

之前所有层中，影响当前层中该元素计算的，所有元素

特征图计算公式

$\frac{h+2padding-k_{size}}{stride}+1$
$\frac{w+2padding-k_{size}}{stride}+1$

卷积核

多个卷积核的旨在实现对不同特征的响应，并实现特征之间的相关性分析

池化层

降低卷积层对位置的敏感性
降低对空间降采样表示的敏感性

特殊方法

BatchNormalization 批规范化

https://blog.csdn.net/L_fengzifei/article/details/124949093

网络模型

• VGG

block概念的提出，简化模型够早
小卷积核更有效

• Inception

聚合不同的卷积层，提取不同的特征

• resnet – 还要再看
  关于残差块的解释
  从复杂度的角度分析：https://zh-v2.d2l.ai/chapter_convolutional-modern/resnet.html
  从恒等映射的角度分析：https://zh-v2.d2l.ai/chapter_convolutional-modern/resnet.html

• 残差块的简洁实现

class ResidualBlock(nn.Module):
	def __init__(self,input_channels,num_channels,use_1x1conv=False,strides=1)
	super(ResidualBlock,self).__init__():
	self.conv1=nn.Conv2d(input_channels,num_channels,kernel_size=3,padding=1,stride=strides)
	self.conv2=nn.Conv2d(input_channels,num_channels,kernel_size=3,padding=1)
	if use_1x1conv:
		self.conv3=nn.Conv2d(input_channels,num_channels,kernel_size=1,stride=strides)
	else:
		self.conv3=None

	self.bn1=nn.BatchNorm2d(num_channels)
	self.bn2=nn.BatchNorm2d(num_channels)

	def forward(self,x):
		Y=F.relu(self.bn1(self.conv1(x)))
		Y=self.bn2(self.conv2(Y))
		if self.conv3:
			x=self.conv3(x)
		Y+=x
		return F.relu(Y)

def resnet_block(input_channels,num_channels,num_residuals,first_block=False):
	blk=[]
	for i in range(num_residuals):
		if i==0 and not first_block:
			blk.append(Residual(input_channels,num_channels,use_1x1conv=True,strides=2))
		else:
			blk.append(Residual(num_channels,num_channels))
	return blk

b1=nn.Sequential(nn.Conv2d(1,64,kernel_size=7,stride=2,padding=3),
				nn.BatchNorm2d(64),nn.Relu(),
				nn.MaxPool2d(kernel_size=3,stride=2,padding=1))
b2=nn.Sequential(*residual_block(64,64,2,first_block=True))
b3=nn.Sequential(*residual_block(64,128,2))
b4=nn.Sequential(*residual_block(128,256,2))
b5=nn.Sequential(*residual_block(256,512,2))

net=nn.Sequential(b1,b2,b3,b4,b5,
				  nn.AdaptiveAvgPool2d(1,1),
				  nn.Flatten(),nn.Linear(512,10))

• DenseNet

def conv_block(input_channels,num_channels):
	return nn.Sequential(nn.BatchNorm2d(input_channels),nn.Relu(),
						 nn.Conv2d(input_channels,num_channels,kernel_size=3,padding=1))

# 当个dense块
class DenseBlock(nn.Module):
	def __init__(slef,num_convs,input_channels,num_channels):
		super(DenseBlock,self).__init__()
		layer=[]
		for i in range(num_convs):
			layer.append(conv_block(num_channels*i+input_channels,num_channels))
		self.net=nn.Sequential(*layer)
	
	def forward(self,x):
		for blk in self.net:
			Y=blk(x)
			x=torch.cat((x,Y),dim=1)
		return x

def transition_block(input_channels,num_channels):
	return nn.Sequential(nn.BatchNorm2d(input_channels),nn.Relu(),
						nn.Conv2d(input_channels,num_channels,kernel_size=1),
						nn.Avgpool2d(kernel_size=2,stride=2))

数据增强

• imgaug https://github.com/aleju/imgaug
• albumentations https://albumentations.ai/

其他参考
https://www.kaggle.com/code/aleksandradeis/data-augmentation-packages-overview/notebook
https://github.com/aleju/imgaug/issues/406
https://mp.weixin.qq.com/s/tdNlCxmz_s1Wwls2qoQXDQ
https://www.kaggle.com/code/jmcslk/pytorch-classification-model-based-on-imgaug/notebook