南昌有哪些做网站的公司,做化学科普网站的目的,对网站的赏析,促销策划特征提取是数据分析和机器学习中的基本概念#xff0c;是将原始数据转换为更适合分析或建模的格式过程中的关键步骤。特征#xff0c;也称为变量或属性#xff0c;是我们用来进行预测、对对象进行分类或从数据中获取见解的数据点的特定特征或属性。
1.AlexNet
paper#…特征提取是数据分析和机器学习中的基本概念是将原始数据转换为更适合分析或建模的格式过程中的关键步骤。特征也称为变量或属性是我们用来进行预测、对对象进行分类或从数据中获取见解的数据点的特定特征或属性。
1.AlexNet
paperhttps://dl.acm.org/doi/pdf/10.1145/3065386
作者 Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton
显然该网络是按照作者名字命名的,但是现在这个bacbone比较老了性能欠佳
框架
整体结构主要由五个卷积层、三个全连接层构成中间穿插着最大池化、ReLU、Dropout 使用ReLu非线性激活函数 code_Pytorch
class AlexNet(nn.Module):Neural network model consisting of layers propsed by AlexNet paper.def __init__(self, num_classes1000):Define and allocate layers for this neural net.Args:num_classes (int): number of classes to predict with this modelsuper().__init__()# input size should be : (b x 3 x 227 x 227)# The image in the original paper states that width and height are 224 pixels, but# the dimensions after first convolution layer do not lead to 55 x 55.self.net nn.Sequential(nn.Conv2d(in_channels3, out_channels96, kernel_size11, stride4), # (b x 96 x 55 x 55)nn.ReLU(),nn.LocalResponseNorm(size5, alpha0.0001, beta0.75, k2), # section 3.3nn.MaxPool2d(kernel_size3, stride2), # (b x 96 x 27 x 27)nn.Conv2d(96, 256, 5, padding2), # (b x 256 x 27 x 27)nn.ReLU(),nn.LocalResponseNorm(size5, alpha0.0001, beta0.75, k2),nn.MaxPool2d(kernel_size3, stride2), # (b x 256 x 13 x 13)nn.Conv2d(256, 384, 3, padding1), # (b x 384 x 13 x 13)nn.ReLU(),nn.Conv2d(384, 384, 3, padding1), # (b x 384 x 13 x 13)nn.ReLU(),nn.Conv2d(384, 256, 3, padding1), # (b x 256 x 13 x 13)nn.ReLU(),nn.MaxPool2d(kernel_size3, stride2), # (b x 256 x 6 x 6))# classifier is just a name for linear layersself.classifier nn.Sequential(nn.Dropout(p0.5, inplaceTrue),nn.Linear(in_features(256 * 6 * 6), out_features4096),nn.ReLU(),nn.Dropout(p0.5, inplaceTrue),nn.Linear(in_features4096, out_features4096),nn.ReLU(),nn.Linear(in_features4096, out_featuresnum_classes),)self.init_bias() # initialize biasdef init_bias(self):for layer in self.net:if isinstance(layer, nn.Conv2d):nn.init.normal_(layer.weight, mean0, std0.01)nn.init.constant_(layer.bias, 0)# original paper 1 for Conv2d layers 2nd, 4th, and 5th conv layersnn.init.constant_(self.net[4].bias, 1)nn.init.constant_(self.net[10].bias, 1)nn.init.constant_(self.net[12].bias, 1)def forward(self, x):Pass the input through the net.Args:x (Tensor): input tensorReturns:output (Tensor): output tensorx self.net(x)x x.view(-1, 256 * 6 * 6) # reduce the dimensions for linear layer inputreturn self.classifier(x)
2.VGG
paper:https://arxiv.org/abs/1409.1556
作者Karen Simonyan, Andrew Zisserman
超级超级经典的网络从14年到现在还是广泛使用
框架
相比AlexNet而言加深了网络的深度VGG1613层conv3层FC和VGG1916层conv3层FC是指表中的D、E两个模型。 code_vgg_Pytorch Modified from https://github.com/pytorch/vision.gitimport mathimport torch.nn as nn
import torch.nn.init as init__all__ [VGG, vgg11, vgg11_bn, vgg13, vgg13_bn, vgg16, vgg16_bn,vgg19_bn, vgg19,
]class VGG(nn.Module):VGG model def __init__(self, features):super(VGG, self).__init__()self.features featuresself.classifier nn.Sequential(nn.Dropout(),nn.Linear(512, 512),nn.ReLU(True),nn.Dropout(),nn.Linear(512, 512),nn.ReLU(True),nn.Linear(512, 10),)# Initialize weightsfor m in self.modules():if isinstance(m, nn.Conv2d):n m.kernel_size[0] * m.kernel_size[1] * m.out_channelsm.weight.data.normal_(0, math.sqrt(2. / n))m.bias.data.zero_()def forward(self, x):x self.features(x)x x.view(x.size(0), -1)x self.classifier(x)return xdef make_layers(cfg, batch_normFalse):layers []in_channels 3for v in cfg:if v M:layers [nn.MaxPool2d(kernel_size2, stride2)]else:conv2d nn.Conv2d(in_channels, v, kernel_size3, padding1)if batch_norm:layers [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplaceTrue)]else:layers [conv2d, nn.ReLU(inplaceTrue)]in_channels vreturn nn.Sequential(*layers)cfg {A: [64, M, 128, M, 256, 256, M, 512, 512, M, 512, 512, M],B: [64, 64, M, 128, 128, M, 256, 256, M, 512, 512, M, 512, 512, M],D: [64, 64, M, 128, 128, M, 256, 256, 256, M, 512, 512, 512, M, 512, 512, 512, M],E: [64, 64, M, 128, 128, M, 256, 256, 256, 256, M, 512, 512, 512, 512, M, 512, 512, 512, 512, M],
}def vgg11():VGG 11-layer model (configuration A)return VGG(make_layers(cfg[A]))def vgg11_bn():VGG 11-layer model (configuration A) with batch normalizationreturn VGG(make_layers(cfg[A], batch_normTrue))def vgg13():VGG 13-layer model (configuration B)return VGG(make_layers(cfg[B]))def vgg13_bn():VGG 13-layer model (configuration B) with batch normalizationreturn VGG(make_layers(cfg[B], batch_normTrue))def vgg16():VGG 16-layer model (configuration D)return VGG(make_layers(cfg[D]))def vgg16_bn():VGG 16-layer model (configuration D) with batch normalizationreturn VGG(make_layers(cfg[D], batch_normTrue))def vgg19():VGG 19-layer model (configuration E)return VGG(make_layers(cfg[E]))def vgg19_bn():VGG 19-layer model (configuration E) with batch normalizationreturn VGG(make_layers(cfg[E], batch_normTrue))
3.ResNet
paper:https://arxiv.org/abs/1512.03385
作者Kaiming He、Xiangyu Zhang、Shaoqing RenMicrosoft Research
使用残差网络避免模型变深带来的梯度爆炸和梯度消失的问题使得网络层数可以达到很深。
框架
残差连接
1完成恒等映射浅层特征可以直接的传递到深层特征中。
2梯度回传深层的梯度可以通过残差的结构直接传递到浅层的网络中。 基于上面的分析提出残差连接结构构建了不同的网络有18、34、50、101、152等。 code_ResNet_Pytorch
import torch
import torch.nn as nn
import torchvision.models.resnet
from torchvision.models.resnet import BasicBlock, Bottleneckclass ResNet(torchvision.models.resnet.ResNet):def __init__(self, block, layers, num_classes1000, group_normFalse):if group_norm:norm_layer lambda x: nn.GroupNorm(32, x)else:norm_layer Nonesuper(ResNet, self).__init__(block, layers, num_classes, norm_layernorm_layer)if not group_norm:self.maxpool nn.MaxPool2d(kernel_size3, stride2, padding0, ceil_modeTrue) # changefor i in range(2, 5):getattr(self, layer%d%i)[0].conv1.stride (2,2)getattr(self, layer%d%i)[0].conv2.stride (1,1)def resnet18(pretrainedFalse):Constructs a ResNet-18 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(BasicBlock, [2, 2, 2, 2])if pretrained:model.load_state_dict(model_zoo.load_url(model_urls[resnet18]))return modeldef resnet34(pretrainedFalse):Constructs a ResNet-34 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(BasicBlock, [3, 4, 6, 3])if pretrained:model.load_state_dict(model_zoo.load_url(model_urls[resnet34]))return modeldef resnet50(pretrainedFalse):Constructs a ResNet-50 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(Bottleneck, [3, 4, 6, 3])if pretrained:model.load_state_dict(model_zoo.load_url(model_urls[resnet50]))return modeldef resnet50_gn(pretrainedFalse):Constructs a ResNet-50 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(Bottleneck, [3, 4, 6, 3], group_normTrue)if pretrained:model.load_state_dict(model_zoo.load_url(model_urls[resnet50]))return modeldef resnet101(pretrainedFalse):Constructs a ResNet-101 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(Bottleneck, [3, 4, 23, 3])if pretrained:model.load_state_dict(model_zoo.load_url(model_urls[resnet101]))return modeldef resnet101_gn(pretrainedFalse):Constructs a ResNet-101 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(Bottleneck, [3, 4, 23, 3], group_normTrue)return modeldef resnet152(pretrainedFalse):Constructs a ResNet-152 model.Args:pretrained (bool): If True, returns a model pre-trained on ImageNetmodel ResNet(Bottleneck, [3, 8, 36, 3])if pretrained:model.load_state_dict(model_zoo.load_url(model_urls[resnet152]))return model
