第J9周：Inception v3算法实战与解析（pytorch版）
>- **🍨 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]**
📌本周任务：📌
-
了解并学习InceptionV3相对与InceptionV1有哪些改进的地方
-
使用Inception完成天气识别
🏡 我的环境：
- 语言环境：Python3.8
- 编译器：Jupyter Notebook
- 深度学习环境：Pytorch
-
- torch==2.3.1+cu118
-
- torchvision==0.18.1+cu118
一、理论基础
Inception v3论文：Rethinking the Inception Architecture for Computer Vision
Inception v3由谷歌研究员Christian Szegedy等人在2015年的论文《Rethinking the Inception Architecture for Computer Vision》中提出。Inception v3是Inception网络系列的第三个版本，它在ImageNet图像识别竞赛中取得了优异成绩，尤其是在大规模图像识别任务中表现出色。
Inception v3的主要特点如下：
- 更深的网络结构：Inception v3比之前的Inception网络结构更深，包含了48层卷积层。这使得网络可以提取更多层次的特征，从而在图像识别任务上取得更好的效果。
- 使用Factorized Convolutions：Inception v3采用了Factorized Convolutions（分解卷积），将较大的卷积核分解为多个较小的卷积核。这种方法可以降低网络的参数数量，减少计算复杂度，同时保持良好的性能。
- 使用Batch Normalization：Inception v3在每个卷积层之后都添加了Batch Normalization（BN），这有助于网络的收敛和泛化能力。BN可以减少Internal Covariate Shift（内部协变量偏移）现象，加快训练速度，同时提高模型的鲁棒性。
- 辅助分类器：Inception v3引入了辅助分类器，可以在网络训练过程中提供额外的梯度信息，帮助网络更好地学习特征。辅助分类器位于网络的某个中间层，其输出会与主分类器的输出进行加权融合，从而得到最终的预测结果。
- 基于RMSProp的优化器：Inception v3使用了RMSProp优化器进行训练。相比于传统的随机梯度下降（SGD）方法，RMSProp可以自适应地调整学习率，使得训练过程更加稳定，收敛速度更快。
Inception v3在图像分类、物体检测和图像分割等计算机视觉任务中均取得了显著的效果。然而，由于其较大的网络结构和计算复杂度，Inception v3在实际应用中可能需要较高的硬件要求。
相对于Inception v1的Inception Module结构，Inception v3中做出了如下改动：
●将 5×5 的卷积分解为两个 3×3 的卷积运算以提升计算速度。尽管这有点违反直觉，但一个 5×5 的卷积在计算成本上是一个 3×3 卷积的 2.78 倍。所以叠加两个 3×3 卷积实际上在性能上会有所提升，如下图所示：
●此外，作者将 n×n 的卷积核尺寸分解为 1×n 和 n×1 两个卷积。例如，一个 3×3 的卷积等价于首先执行一个 1×3 的卷积再执行一个 3×1 的卷积。他们还发现这种方法在成本上要比单个 3×3 的卷积降低 33%，这一结构如下图所示：
此处如果 n=3，则与上一张图像一致。最左侧的 5x5 卷积可被表示为两个 3x3 卷积，它们又可以被表示为 1x3 和 3x1 卷积。
模块中的滤波器组被扩展（即变得更宽而不是更深），以解决表征性瓶颈。如果该模块没有被拓展宽度，而是变得更深，那么维度会过多减少，造成信息损失。如下图所示：
最后实现的inception v3网络是上图结构图如下：
二、前期准备
1. 设置GPU
如果设备上支持GPU就使用GPU,否则使用CPU
import warnings
warnings.filterwarnings("ignore") #忽略警告信息
import torch
device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
device运行结果：
device(type='cuda')2. 导入数据
import pathlib
data_dir=r'D:\THE MNIST DATABASE\weather_photos'
data_dir=pathlib.Path(data_dir)
img_count=len(list(data_dir.glob('*/*')))
print("图片总数为：",img_count)运行结果：
图片总数为： 11253. 查看数据集分类
data_paths=list(data_dir.glob('*'))
classNames=[str(path).split('\\')[3] for path in data_paths]
classNames运行结果：
['cloudy', 'rain', 'shine', 'sunrise']4. 随机查看图片
随机抽取数据集中的10张图片进行查看
import PIL,random
import matplotlib.pyplot as plt
from PIL import Image
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
data_paths2=list(data_dir.glob('*/*'))
plt.figure(figsize=(20,8))
#plt.suptitle("OreoCC的案例",fontsize=15)
for i in range(10):
plt.subplot(2,5,i+1)
plt.axis("off")
image=random.choice(data_paths2) #随机选择一个图片
plt.title(image.parts[-2],fontsize=20) #通过glob对象取出他的文件夹名称，即分类名
plt.imshow(Image.open(str(image))) #显示图片运行结果：
5. 图片预处理
import torchvision.transforms as transforms
from torchvision import transforms,datasets
train_transforms=transforms.Compose([
transforms.Resize([224,224]), #将图片统一尺寸
transforms.RandomHorizontalFlip(), #将图片随机水平翻转
transforms.RandomRotation(0.2), #将图片按照0.2的弧度值随机旋转
transforms.ToTensor(), #将图片转换为tensor
transforms.Normalize( #标准化处理->转换为正态分布，使模型更容易收敛
mean=[0.485,0.456,0.406],
std=[0.229,0.224,0.225]
)
])
total_data=datasets.ImageFolder(
r'D:\THE MNIST DATABASE\weather_photos',
transform=train_transforms
)
total_data运行结果：
Dataset ImageFolder
Number of datapoints: 1125
Root location: D:\THE MNIST DATABASE\weather_photos
StandardTransform
Transform: Compose(
Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=True)
RandomHorizontalFlip(p=0.5)
RandomRotation(degrees=[-0.2, 0.2], interpolation=nearest, expand=False, fill=0)
ToTensor()
Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
)将数据集分类情况进行映射输出：
total_data.class_to_idx运行结果：
{'cloudy': 0, 'rain': 1, 'shine': 2, 'sunrise': 3}6. 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运行结果：
(<torch.utils.data.dataset.Subset at 0x1f39bf80a90>,
<torch.utils.data.dataset.Subset at 0x1f3bc462210>)查看训练集和测试集的数据数量：
train_size,test_size运行结果：
(900, 225)7. 加载数据集
batch_size=8
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
)查看测试集的情况：
for x,y in train_dl:
print("Shape of x [N,C,H,W]:",x.shape)
print("Shape of y:",y.shape,y.dtype)
break运行结果：
Shape of x [N,C,H,W]: torch.Size([8, 3, 224, 224])
Shape of y: torch.Size([8]) torch.int64三、手动搭建网络模型
1. BasicConv2d模块
import torch.nn as nn
import torch.nn.functional as F
class BasicConv2d(nn.Module):
def __init__(self,in_channels,out_channels,**kwargs):
super(BasicConv2d,self).__init__()
self.conv=nn.Conv2d(in_channels,out_channels,bias=False,**kwargs)
self.bn=nn.BatchNorm2d(out_channels,eps=0.001)
self.relu=nn.ReLU(inplace=True)
def forward(self,x):
x=self.conv(x)
x=self.bn(x)
x=self.relu(x)
return x2. Inception-A
class InceptionA(nn.Module):
def __init__(self,in_channels,pool_features):
super(InceptionA,self).__init__()
self.branch1x1=BasicConv2d(in_channels,64,kernel_size=1)
self.branch5x5_1=BasicConv2d(in_channels,48,kernel_size=1)
self.branch5x5_2=BasicConv2d(48,64,kernel_size=5,padding=2)
self.branch3x3dbl_1=BasicConv2d(in_channels,64,kernel_size=1)
self.branch3x3dbl_2=BasicConv2d(64,96,kernel_size=3,padding=1)
self.branch3x3dbl_3=BasicConv2d(96,96,kernel_size=3,padding=1)
self.branch_pool=BasicConv2d(in_channels,pool_features,kernel_size=1)
def forward(self,x):
branch1x1=self.branch1x1(x)
branch5x5=self.branch5x5_1(x)
branch5x5=self.branch5x5_2(branch5x5)
branch3x3dbl=self.branch3x3dbl_1(x)
branch3x3dbl=self.branch3x3dbl_2(branch3x3dbl)
branch3x3dbl=self.branch3x3dbl_3(branch3x3dbl)
branch_pool=F.avg_pool2d(x,kernel_size=3,stride=1,padding=1)
branch_pool=self.branch_pool(branch_pool)
outputs=[branch1x1,branch5x5,branch3x3dbl,branch_pool]
return torch.cat(outputs,1)InceptionA模块包含四个分支，每个分支使用不同的卷积核大小和参数。其中branch1x1、branch5x5_2、branch3x3dbl_3使用较小的卷积核，可以减少参数数量和计算量，提高网络效率。branch_pool使用平均池化的方式进行特征提取和降维。最后，将四个分支的结果在通道维度上进行拼接，输出InceptionA的结果。
3. Inception-B
class InceptionB(nn.Module):
def __init__(self,in_channels,channels_7x7):
super(InceptionB,self).__init__()
self.branch1x1=BasicConv2d(in_channels,192,kernel_size=1)
c7=channels_7x7
self.branch7x7_1=BasicConv2d(in_channels,c7,kernel_size=1)
self.branch7x7_2=BasicConv2d(c7,c7,kernel_size=(1,7),padding=(0,3))
self.branch7x7_3=BasicConv2d(c7,192,kernel_size=(7,1),padding=(3,0))
self.branch7x7dbl_1=BasicConv2d(in_channels,c7,kernel_size=1)
self.branch7x7dbl_2=BasicConv2d(c7,c7,kernel_size=(7,1),padding=(3,0))
self.branch7x7dbl_3=BasicConv2d(c7,c7,kernel_size=(1,7),padding=(0,3))
self.branch7x7dbl_4=BasicConv2d(c7,c7,kernel_size=(7,1),padding=(3,0))
self.branch7x7dbl_5=BasicConv2d(c7,192,kernel_size=(1,7),padding=(0,3))
self.branch_pool=BasicConv2d(in_channels,192,kernel_size=1)
def forward(self,x):
branch1x1=self.branch1x1(x)
branch7x7=self.branch7x7_1(x)
branch7x7=self.branch7x7_2(branch7x7)
branch7x7=self.branch7x7_3(branch7x7)
branch7x7dbl=self.branch7x7dbl_1(x)
branch7x7dbl=self.branch7x7dbl_2(branch7x7dbl)
branch7x7dbl=self.branch7x7dbl_3(branch7x7dbl)
branch7x7dbl=self.branch7x7dbl_4(branch7x7dbl)
branch7x7dbl=self.branch7x7dbl_5(branch7x7dbl)
branch_pool=F.avg_pool2d(x,kernel_size=3,stride=1,padding=1)
branch_pool=self.branch_pool(branch_pool)
outputs=[branch1x1,branch7x7,branch7x7dbl,branch_pool]
return torch.cat(outputs,1)InceptionB模块包含四个分支，其中branch7x7_2和branch7x7_3使用不同大小的卷积核进行多次卷积，可以提高特征的表达能力。branch7x7dbl_2、branch7x7dbl_3、branch7x7dbl_4、branch7x7dbl_5也类似地使用多个不同大小的卷积核进行多次卷积，提高了特征的表达能力，并且较好地保留了空间尺寸。branch_pool仍然使用平均池化的方式进行特征提取和降维。最后，将四个分支的结果在通道维度上进行拼接，输出InceptionB的结果。
4. Inception-C
class InceptionC(nn.Module):
def __init__(self,in_channels):
super(InceptionC,self).__init__()
self.branch1x1=BasicConv2d(in_channels,320,kernel_size=1)
self.branch3x3_1=BasicConv2d(in_channels,384,kernel_size=1)
self.branch3x3_2a=BasicConv2d(384,384,kernel_size=(1,3),padding=(0,1))
self.branch3x3_2b=BasicConv2d(384,384,kernel_size=(3,1),padding=(1,0))
self.branch3x3dbl_1=BasicConv2d(in_channels,448,kernel_size=1)
self.branch3x3dbl_2=BasicConv2d(448,384,kernel_size=3,padding=1)
self.branch3x3dbl_3a=BasicConv2d(384,384,kernel_size=(1,3),padding=(0,1))
self.branch3x3dbl_3b=BasicConv2d(384,384,kernel_size=(3,1),padding=(1,0))
self.branch_pool=BasicConv2d(in_channels,192,kernel_size=1)
def forward(self,x):
branch1x1=self.branch1x1(x)
branch3x3=self.branch3x3_1(x)
branch3x3=[self.branch3x3_2a(branch3x3),
self.branch3x3_2b(branch3x3),]
branch3x3=torch.cat(branch3x3,1)
branch3x3dbl=self.branch3x3dbl_1(x)
branch3x3dbl=self.branch3x3dbl_2(branch3x3dbl)
branch3x3dbl=[self.branch3x3dbl_3a(branch3x3dbl),
self.branch3x3dbl_3b(branch3x3dbl),]
branch3x3dbl=torch.cat(branch3x3dbl,1)
branch_pool=F.avg_pool2d(x,kernel_size=3,stride=1,padding=1)
branch_pool=self.branch_pool(branch_pool)
outputs=[branch1x1,branch3x3,branch3x3dbl,branch_pool]
return torch.cat(outputs,1)InceptionC模块同样包含三个分支，其中branch3x3_2b和branch3x3dbl_2使用大小为3x3的卷积核，并且padding=1，dilation=1，可以一定程度上扩大感受野。这两个分支多次卷积，可以提高特征的表达能力。branch_pool仍然使用平均池化的方式进行特征提取和降维。最后，将三个分支的结果在通道维度上进行拼接，输出InceptionB的结果。
5、Reduction-A
class ReductionA(nn.Module):
def __init__(self,in_channels):
super(ReductionA,self).__init__()
self.branch3x3=BasicConv2d(in_channels,384,kernel_size=3,stride=2)
self.branch3x3dbl_1=BasicConv2d(in_channels,64,kernel_size=1)
self.branch3x3dbl_2=BasicConv2d(64,96,kernel_size=3,padding=1)
self.branch3x3dbl_3=BasicConv2d(96,96,kernel_size=3,stride=2)
def forward(self,x):
branch3x3=self.branch3x3(x)
branch3x3dbl=self.branch3x3dbl_1(x)
branch3x3dbl=self.branch3x3dbl_2(branch3x3dbl)
branch3x3dbl=self.branch3x3dbl_3(branch3x3dbl)
branch_pool=F.max_pool2d(x,kernel_size=3,stride=2)
outputs=[branch3x3,branch3x3dbl,branch_pool]
return torch.cat(outputs,1)ReductionA模块包含三个分支，其中branch3x3_1使用3x3的卷积核进行卷积，通过stride=2来降维，同时提取特征。branch3x3_2a、branch3x3_2b、branch3x3_2c使用多层卷积对特征进行提取和表达，同时通过stride=2来降维和压缩特征，减少计算量。branch_pool使用max pooling的方式进行特征提取和降维，与其他模块类似。最后，将三个分支的结果在通道维度上进行拼接，输出ReductionA的结果。
6、Reduction-B
class ReductionB(nn.Module):
def __init__(self,in_channels):
super(ReductionB,self).__init__()
self.branch3x3_1=BasicConv2d(in_channels,192,kernel_size=1)
self.branch3x3_2=BasicConv2d(192,320,kernel_size=3,stride=2)
self.branch7x7x3_1=BasicConv2d(in_channels,192,kernel_size=1)
self.branch7x7x3_2=BasicConv2d(192,192,kernel_size=(1,7),padding=(0,3))
self.branch7x7x3_3=BasicConv2d(192,192,kernel_size=(7,1),padding=(3,0))
self.branch7x7x3_4=BasicConv2d(192,192,kernel_size=3,stride=2)
def forward(self,x):
branch3x3=self.branch3x3_1(x)
branch3x3=self.branch3x3_2(branch3x3)
branch7x7x3=self.branch7x7x3_1(x)
branch7x7x3=self.branch7x7x3_2(branch7x7x3)
branch7x7x3=self.branch7x7x3_3(branch7x7x3)
branch7x7x3=self.branch7x7x3_4(branch7x7x3)
branch_pool=F.max_pool2d(x,kernel_size=3,stride=2)
outputs=[branch3x3,branch7x7x3,branch_pool]
return torch.cat(outputs,1)7、辅助分支
class InceptionAux(nn.Module):
def __init__(self,in_channels,num_classes):
super(InceptionAux,self).__init__()
self.conv0=BasicConv2d(in_channels,128,kernel_size=1)
self.conv1=BasicConv2d(128,768,kernel_size=5)
self.conv1.stddev=0.01
self.fc=nn.Linear(768,num_classes)
self.fc.stddev=0.001
def forward(self,x):
#17x17x768
x=F.avg_pool2d(x,kernel_size=5,stride=3)
# 5x5x768
x=self.conv0(x)
# 5x5x128
x=self.conv1(x)
# 1x1x768
x=x.view(x.size(0),-1)
# 768
x=self.fc(x)
# 1000
return x8、InceptionV3实现
class InceptionV3(nn.Module):
def __init__(self,num_classes=1000,aux_logits=False,transform_input=False):
super(InceptionV3,self).__init__()
self.aux_logits=aux_logits
self.transform_input=transform_input
self.Conv2d_1a_3x3=BasicConv2d(3,32,kernel_size=3,stride=2)
self.Conv2d_2a_3x3=BasicConv2d(32,32,kernel_size=3)
self.Conv2d_2b_3x3=BasicConv2d(32,64,kernel_size=3,padding=1)
self.Conv2d_3b_1x1=BasicConv2d(64,80,kernel_size=1)
self.Conv2d_4a_3x3=BasicConv2d(80,192,kernel_size=3)
self.Mixed_5b=InceptionA(192,pool_features=32)
self.Mixed_5c=InceptionA(256,pool_features=64)
self.Mixed_5d=InceptionA(288,pool_features=64)
self.Mixed_6a=ReductionA(288)
self.Mixed_6b=InceptionB(768,channels_7x7=128)
self.Mixed_6c=InceptionB(768,channels_7x7=160)
self.Mixed_6d=InceptionB(768,channels_7x7=160)
self.Mixed_6e=InceptionB(768,channels_7x7=192)
if aux_logits:
self.AuxLogits=InceptionAux(768,num_classes)
self.Mixed_7a=ReductionB(768)
self.Mixed_7b=InceptionC(1280)
self.Mixed_7c=InceptionC(2048)
self.fc=nn.Linear(2048,num_classes)
def forward(self,x):
if self.transform_input:
x=x.clone()
x[:,0]=x[:,0]*(0.229/0.5)+(0.485-0.5)/0.5
x[:,1]=x[:,1]*(0.224/0.5)+(0.456-0.5)/0.5
x[:,2]=x[:,2]*(0.225/0.5)+(0.406-0.5)/0.5
# 229x229x3
x=self.Conv2d_1a_3x3(x)
# 149x149x32
x=self.Conv2d_2a_3x3(x)
# 147x147x32
x=self.Conv2d_2b_3x3(x)
# 147x147x64
x=F.max_pool2d(x,kernel_size=3,stride=2)
# 73x73x64
x=self.Conv2d_3b_1x1(x)
# 73x73x80
x=self.Conv2d_4a_3x3(x)
# 71x71x192
x=F.max_pool2d(x,kernel_size=3,stride=2)
# 35x35x192
x=self.Mixed_5b(x)
# 35x35x256
x=self.Mixed_5c(x)
# 35x35x288
x=self.Mixed_5d(x)
# 35x35x288
x=self.Mixed_6a(x)
# 17x17x768
x=self.Mixed_6b(x)
# 17x17x768
x=self.Mixed_6c(x)
# 17x17x768
x=self.Mixed_6d(x)
# 17x17x768
x=self.Mixed_6e(x)
# 17x17x768
if self.training and self.aux_logits:
aux=self.AuxLogits(x)
# 17x17x768
x=self.Mixed_7a(x)
# 8x8x1280
x=self.Mixed_7b(x)
# 8x8x2048
x=self.Mixed_7c(x)
# 8x8x2048
x=F.avg_pool2d(x,kernel_size=5)
# 1x1x2048
x=F.dropout(x,training=self.training)
# 1x1x2048
x=x.view(x.size(0),-1)
# 2048
x=self.fc(x)
# 1000(num_classes)
if self.training and self.aux_logits:
return x,aux
return xmodel=InceptionV3(num_classes=4).to(device)
model运行结果：
InceptionV3(
(Conv2d_1a_3x3): BasicConv2d(
(conv): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(Conv2d_2a_3x3): BasicConv2d(
(conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(Conv2d_2b_3x3): BasicConv2d(
(conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(Conv2d_3b_1x1): BasicConv2d(
(conv): Conv2d(64, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(80, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(Conv2d_4a_3x3): BasicConv2d(
(conv): Conv2d(80, 192, kernel_size=(3, 3), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(Mixed_5b): InceptionA(
(branch1x1): BasicConv2d(
(conv): Conv2d(192, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch5x5_1): BasicConv2d(
(conv): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(48, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch5x5_2): BasicConv2d(
(conv): Conv2d(48, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_1): BasicConv2d(
(conv): Conv2d(192, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_2): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(192, 32, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_5c): InceptionA(
(branch1x1): BasicConv2d(
(conv): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch5x5_1): BasicConv2d(
(conv): Conv2d(256, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(48, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch5x5_2): BasicConv2d(
(conv): Conv2d(48, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_1): BasicConv2d(
(conv): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_2): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_5d): InceptionA(
(branch1x1): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch5x5_1): BasicConv2d(
(conv): Conv2d(288, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(48, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch5x5_2): BasicConv2d(
(conv): Conv2d(48, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_1): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_2): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_6a): ReductionA(
(branch3x3): BasicConv2d(
(conv): Conv2d(288, 384, kernel_size=(3, 3), stride=(2, 2), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_1): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_2): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), bias=False)
(bn): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_6b): InceptionB(
(branch1x1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_1): BasicConv2d(
(conv): Conv2d(768, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_2): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_3): BasicConv2d(
(conv): Conv2d(128, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_1): BasicConv2d(
(conv): Conv2d(768, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_2): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_3): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_4): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_5): BasicConv2d(
(conv): Conv2d(128, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_6c): InceptionB(
(branch1x1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_1): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_2): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_3): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_1): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_2): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_3): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_4): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_5): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_6d): InceptionB(
(branch1x1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_1): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_2): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_3): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_1): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_2): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_3): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_4): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_5): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_6e): InceptionB(
(branch1x1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_2): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7_3): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_2): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_3): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_4): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7dbl_5): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_7a): ReductionB(
(branch3x3_1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_2): BasicConv2d(
(conv): Conv2d(192, 320, kernel_size=(3, 3), stride=(2, 2), bias=False)
(bn): BatchNorm2d(320, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7x3_1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7x3_2): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7x3_3): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch7x7x3_4): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(3, 3), stride=(2, 2), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_7b): InceptionC(
(branch1x1): BasicConv2d(
(conv): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(320, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_1): BasicConv2d(
(conv): Conv2d(1280, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_2a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_2b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_1): BasicConv2d(
(conv): Conv2d(1280, 448, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(448, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_2): BasicConv2d(
(conv): Conv2d(448, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(1280, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(Mixed_7c): InceptionC(
(branch1x1): BasicConv2d(
(conv): Conv2d(2048, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(320, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_1): BasicConv2d(
(conv): Conv2d(2048, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_2a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3_2b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_1): BasicConv2d(
(conv): Conv2d(2048, 448, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(448, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_2): BasicConv2d(
(conv): Conv2d(448, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch3x3dbl_3b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(bn): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(branch_pool): BasicConv2d(
(conv): Conv2d(2048, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(fc): Linear(in_features=2048, out_features=4, bias=True)
)9. 查看模型详情
import torchsummary as summary
summary.summary(model,(3,299,299))运行结果：
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 32, 149, 149] 864
BatchNorm2d-2 [-1, 32, 149, 149] 64
ReLU-3 [-1, 32, 149, 149] 0
BasicConv2d-4 [-1, 32, 149, 149] 0
Conv2d-5 [-1, 32, 147, 147] 9,216
BatchNorm2d-6 [-1, 32, 147, 147] 64
ReLU-7 [-1, 32, 147, 147] 0
BasicConv2d-8 [-1, 32, 147, 147] 0
Conv2d-9 [-1, 64, 147, 147] 18,432
BatchNorm2d-10 [-1, 64, 147, 147] 128
ReLU-11 [-1, 64, 147, 147] 0
BasicConv2d-12 [-1, 64, 147, 147] 0
Conv2d-13 [-1, 80, 73, 73] 5,120
BatchNorm2d-14 [-1, 80, 73, 73] 160
ReLU-15 [-1, 80, 73, 73] 0
BasicConv2d-16 [-1, 80, 73, 73] 0
Conv2d-17 [-1, 192, 71, 71] 138,240
BatchNorm2d-18 [-1, 192, 71, 71] 384
ReLU-19 [-1, 192, 71, 71] 0
BasicConv2d-20 [-1, 192, 71, 71] 0
Conv2d-21 [-1, 64, 35, 35] 12,288
BatchNorm2d-22 [-1, 64, 35, 35] 128
ReLU-23 [-1, 64, 35, 35] 0
BasicConv2d-24 [-1, 64, 35, 35] 0
Conv2d-25 [-1, 48, 35, 35] 9,216
BatchNorm2d-26 [-1, 48, 35, 35] 96
ReLU-27 [-1, 48, 35, 35] 0
BasicConv2d-28 [-1, 48, 35, 35] 0
Conv2d-29 [-1, 64, 35, 35] 76,800
BatchNorm2d-30 [-1, 64, 35, 35] 128
ReLU-31 [-1, 64, 35, 35] 0
BasicConv2d-32 [-1, 64, 35, 35] 0
Conv2d-33 [-1, 64, 35, 35] 12,288
BatchNorm2d-34 [-1, 64, 35, 35] 128
ReLU-35 [-1, 64, 35, 35] 0
BasicConv2d-36 [-1, 64, 35, 35] 0
Conv2d-37 [-1, 96, 35, 35] 55,296
BatchNorm2d-38 [-1, 96, 35, 35] 192
ReLU-39 [-1, 96, 35, 35] 0
BasicConv2d-40 [-1, 96, 35, 35] 0
Conv2d-41 [-1, 96, 35, 35] 82,944
BatchNorm2d-42 [-1, 96, 35, 35] 192
ReLU-43 [-1, 96, 35, 35] 0
BasicConv2d-44 [-1, 96, 35, 35] 0
Conv2d-45 [-1, 32, 35, 35] 6,144
BatchNorm2d-46 [-1, 32, 35, 35] 64
ReLU-47 [-1, 32, 35, 35] 0
BasicConv2d-48 [-1, 32, 35, 35] 0
InceptionA-49 [-1, 256, 35, 35] 0
Conv2d-50 [-1, 64, 35, 35] 16,384
BatchNorm2d-51 [-1, 64, 35, 35] 128
ReLU-52 [-1, 64, 35, 35] 0
BasicConv2d-53 [-1, 64, 35, 35] 0
Conv2d-54 [-1, 48, 35, 35] 12,288
BatchNorm2d-55 [-1, 48, 35, 35] 96
ReLU-56 [-1, 48, 35, 35] 0
BasicConv2d-57 [-1, 48, 35, 35] 0
Conv2d-58 [-1, 64, 35, 35] 76,800
BatchNorm2d-59 [-1, 64, 35, 35] 128
ReLU-60 [-1, 64, 35, 35] 0
BasicConv2d-61 [-1, 64, 35, 35] 0
Conv2d-62 [-1, 64, 35, 35] 16,384
BatchNorm2d-63 [-1, 64, 35, 35] 128
ReLU-64 [-1, 64, 35, 35] 0
BasicConv2d-65 [-1, 64, 35, 35] 0
Conv2d-66 [-1, 96, 35, 35] 55,296
BatchNorm2d-67 [-1, 96, 35, 35] 192
ReLU-68 [-1, 96, 35, 35] 0
BasicConv2d-69 [-1, 96, 35, 35] 0
Conv2d-70 [-1, 96, 35, 35] 82,944
BatchNorm2d-71 [-1, 96, 35, 35] 192
ReLU-72 [-1, 96, 35, 35] 0
BasicConv2d-73 [-1, 96, 35, 35] 0
Conv2d-74 [-1, 64, 35, 35] 16,384
BatchNorm2d-75 [-1, 64, 35, 35] 128
ReLU-76 [-1, 64, 35, 35] 0
BasicConv2d-77 [-1, 64, 35, 35] 0
InceptionA-78 [-1, 288, 35, 35] 0
Conv2d-79 [-1, 64, 35, 35] 18,432
BatchNorm2d-80 [-1, 64, 35, 35] 128
ReLU-81 [-1, 64, 35, 35] 0
BasicConv2d-82 [-1, 64, 35, 35] 0
Conv2d-83 [-1, 48, 35, 35] 13,824
BatchNorm2d-84 [-1, 48, 35, 35] 96
ReLU-85 [-1, 48, 35, 35] 0
BasicConv2d-86 [-1, 48, 35, 35] 0
Conv2d-87 [-1, 64, 35, 35] 76,800
BatchNorm2d-88 [-1, 64, 35, 35] 128
ReLU-89 [-1, 64, 35, 35] 0
BasicConv2d-90 [-1, 64, 35, 35] 0
Conv2d-91 [-1, 64, 35, 35] 18,432
BatchNorm2d-92 [-1, 64, 35, 35] 128
ReLU-93 [-1, 64, 35, 35] 0
BasicConv2d-94 [-1, 64, 35, 35] 0
Conv2d-95 [-1, 96, 35, 35] 55,296
BatchNorm2d-96 [-1, 96, 35, 35] 192
ReLU-97 [-1, 96, 35, 35] 0
BasicConv2d-98 [-1, 96, 35, 35] 0
Conv2d-99 [-1, 96, 35, 35] 82,944
BatchNorm2d-100 [-1, 96, 35, 35] 192
ReLU-101 [-1, 96, 35, 35] 0
BasicConv2d-102 [-1, 96, 35, 35] 0
Conv2d-103 [-1, 64, 35, 35] 18,432
BatchNorm2d-104 [-1, 64, 35, 35] 128
ReLU-105 [-1, 64, 35, 35] 0
BasicConv2d-106 [-1, 64, 35, 35] 0
InceptionA-107 [-1, 288, 35, 35] 0
Conv2d-108 [-1, 384, 17, 17] 995,328
BatchNorm2d-109 [-1, 384, 17, 17] 768
ReLU-110 [-1, 384, 17, 17] 0
BasicConv2d-111 [-1, 384, 17, 17] 0
Conv2d-112 [-1, 64, 35, 35] 18,432
BatchNorm2d-113 [-1, 64, 35, 35] 128
ReLU-114 [-1, 64, 35, 35] 0
BasicConv2d-115 [-1, 64, 35, 35] 0
Conv2d-116 [-1, 96, 35, 35] 55,296
BatchNorm2d-117 [-1, 96, 35, 35] 192
ReLU-118 [-1, 96, 35, 35] 0
BasicConv2d-119 [-1, 96, 35, 35] 0
Conv2d-120 [-1, 96, 17, 17] 82,944
BatchNorm2d-121 [-1, 96, 17, 17] 192
ReLU-122 [-1, 96, 17, 17] 0
BasicConv2d-123 [-1, 96, 17, 17] 0
ReductionA-124 [-1, 768, 17, 17] 0
Conv2d-125 [-1, 192, 17, 17] 147,456
BatchNorm2d-126 [-1, 192, 17, 17] 384
ReLU-127 [-1, 192, 17, 17] 0
BasicConv2d-128 [-1, 192, 17, 17] 0
Conv2d-129 [-1, 128, 17, 17] 98,304
BatchNorm2d-130 [-1, 128, 17, 17] 256
ReLU-131 [-1, 128, 17, 17] 0
BasicConv2d-132 [-1, 128, 17, 17] 0
Conv2d-133 [-1, 128, 17, 17] 114,688
BatchNorm2d-134 [-1, 128, 17, 17] 256
ReLU-135 [-1, 128, 17, 17] 0
BasicConv2d-136 [-1, 128, 17, 17] 0
Conv2d-137 [-1, 192, 17, 17] 172,032
BatchNorm2d-138 [-1, 192, 17, 17] 384
ReLU-139 [-1, 192, 17, 17] 0
BasicConv2d-140 [-1, 192, 17, 17] 0
Conv2d-141 [-1, 128, 17, 17] 98,304
BatchNorm2d-142 [-1, 128, 17, 17] 256
ReLU-143 [-1, 128, 17, 17] 0
BasicConv2d-144 [-1, 128, 17, 17] 0
Conv2d-145 [-1, 128, 17, 17] 114,688
BatchNorm2d-146 [-1, 128, 17, 17] 256
ReLU-147 [-1, 128, 17, 17] 0
BasicConv2d-148 [-1, 128, 17, 17] 0
Conv2d-149 [-1, 128, 17, 17] 114,688
BatchNorm2d-150 [-1, 128, 17, 17] 256
ReLU-151 [-1, 128, 17, 17] 0
BasicConv2d-152 [-1, 128, 17, 17] 0
Conv2d-153 [-1, 128, 17, 17] 114,688
BatchNorm2d-154 [-1, 128, 17, 17] 256
ReLU-155 [-1, 128, 17, 17] 0
BasicConv2d-156 [-1, 128, 17, 17] 0
Conv2d-157 [-1, 192, 17, 17] 172,032
BatchNorm2d-158 [-1, 192, 17, 17] 384
ReLU-159 [-1, 192, 17, 17] 0
BasicConv2d-160 [-1, 192, 17, 17] 0
Conv2d-161 [-1, 192, 17, 17] 147,456
BatchNorm2d-162 [-1, 192, 17, 17] 384
ReLU-163 [-1, 192, 17, 17] 0
BasicConv2d-164 [-1, 192, 17, 17] 0
InceptionB-165 [-1, 768, 17, 17] 0
Conv2d-166 [-1, 192, 17, 17] 147,456
BatchNorm2d-167 [-1, 192, 17, 17] 384
ReLU-168 [-1, 192, 17, 17] 0
BasicConv2d-169 [-1, 192, 17, 17] 0
Conv2d-170 [-1, 160, 17, 17] 122,880
BatchNorm2d-171 [-1, 160, 17, 17] 320
ReLU-172 [-1, 160, 17, 17] 0
BasicConv2d-173 [-1, 160, 17, 17] 0
Conv2d-174 [-1, 160, 17, 17] 179,200
BatchNorm2d-175 [-1, 160, 17, 17] 320
ReLU-176 [-1, 160, 17, 17] 0
BasicConv2d-177 [-1, 160, 17, 17] 0
Conv2d-178 [-1, 192, 17, 17] 215,040
BatchNorm2d-179 [-1, 192, 17, 17] 384
ReLU-180 [-1, 192, 17, 17] 0
BasicConv2d-181 [-1, 192, 17, 17] 0
Conv2d-182 [-1, 160, 17, 17] 122,880
BatchNorm2d-183 [-1, 160, 17, 17] 320
ReLU-184 [-1, 160, 17, 17] 0
BasicConv2d-185 [-1, 160, 17, 17] 0
Conv2d-186 [-1, 160, 17, 17] 179,200
BatchNorm2d-187 [-1, 160, 17, 17] 320
ReLU-188 [-1, 160, 17, 17] 0
BasicConv2d-189 [-1, 160, 17, 17] 0
Conv2d-190 [-1, 160, 17, 17] 179,200
BatchNorm2d-191 [-1, 160, 17, 17] 320
ReLU-192 [-1, 160, 17, 17] 0
BasicConv2d-193 [-1, 160, 17, 17] 0
Conv2d-194 [-1, 160, 17, 17] 179,200
BatchNorm2d-195 [-1, 160, 17, 17] 320
ReLU-196 [-1, 160, 17, 17] 0
BasicConv2d-197 [-1, 160, 17, 17] 0
Conv2d-198 [-1, 192, 17, 17] 215,040
BatchNorm2d-199 [-1, 192, 17, 17] 384
ReLU-200 [-1, 192, 17, 17] 0
BasicConv2d-201 [-1, 192, 17, 17] 0
Conv2d-202 [-1, 192, 17, 17] 147,456
BatchNorm2d-203 [-1, 192, 17, 17] 384
ReLU-204 [-1, 192, 17, 17] 0
BasicConv2d-205 [-1, 192, 17, 17] 0
InceptionB-206 [-1, 768, 17, 17] 0
Conv2d-207 [-1, 192, 17, 17] 147,456
BatchNorm2d-208 [-1, 192, 17, 17] 384
ReLU-209 [-1, 192, 17, 17] 0
BasicConv2d-210 [-1, 192, 17, 17] 0
Conv2d-211 [-1, 160, 17, 17] 122,880
BatchNorm2d-212 [-1, 160, 17, 17] 320
ReLU-213 [-1, 160, 17, 17] 0
BasicConv2d-214 [-1, 160, 17, 17] 0
Conv2d-215 [-1, 160, 17, 17] 179,200
BatchNorm2d-216 [-1, 160, 17, 17] 320
ReLU-217 [-1, 160, 17, 17] 0
BasicConv2d-218 [-1, 160, 17, 17] 0
Conv2d-219 [-1, 192, 17, 17] 215,040
BatchNorm2d-220 [-1, 192, 17, 17] 384
ReLU-221 [-1, 192, 17, 17] 0
BasicConv2d-222 [-1, 192, 17, 17] 0
Conv2d-223 [-1, 160, 17, 17] 122,880
BatchNorm2d-224 [-1, 160, 17, 17] 320
ReLU-225 [-1, 160, 17, 17] 0
BasicConv2d-226 [-1, 160, 17, 17] 0
Conv2d-227 [-1, 160, 17, 17] 179,200
BatchNorm2d-228 [-1, 160, 17, 17] 320
ReLU-229 [-1, 160, 17, 17] 0
BasicConv2d-230 [-1, 160, 17, 17] 0
Conv2d-231 [-1, 160, 17, 17] 179,200
BatchNorm2d-232 [-1, 160, 17, 17] 320
ReLU-233 [-1, 160, 17, 17] 0
BasicConv2d-234 [-1, 160, 17, 17] 0
Conv2d-235 [-1, 160, 17, 17] 179,200
BatchNorm2d-236 [-1, 160, 17, 17] 320
ReLU-237 [-1, 160, 17, 17] 0
BasicConv2d-238 [-1, 160, 17, 17] 0
Conv2d-239 [-1, 192, 17, 17] 215,040
BatchNorm2d-240 [-1, 192, 17, 17] 384
ReLU-241 [-1, 192, 17, 17] 0
BasicConv2d-242 [-1, 192, 17, 17] 0
Conv2d-243 [-1, 192, 17, 17] 147,456
BatchNorm2d-244 [-1, 192, 17, 17] 384
ReLU-245 [-1, 192, 17, 17] 0
BasicConv2d-246 [-1, 192, 17, 17] 0
InceptionB-247 [-1, 768, 17, 17] 0
Conv2d-248 [-1, 192, 17, 17] 147,456
BatchNorm2d-249 [-1, 192, 17, 17] 384
ReLU-250 [-1, 192, 17, 17] 0
BasicConv2d-251 [-1, 192, 17, 17] 0
Conv2d-252 [-1, 192, 17, 17] 147,456
BatchNorm2d-253 [-1, 192, 17, 17] 384
ReLU-254 [-1, 192, 17, 17] 0
BasicConv2d-255 [-1, 192, 17, 17] 0
Conv2d-256 [-1, 192, 17, 17] 258,048
BatchNorm2d-257 [-1, 192, 17, 17] 384
ReLU-258 [-1, 192, 17, 17] 0
BasicConv2d-259 [-1, 192, 17, 17] 0
Conv2d-260 [-1, 192, 17, 17] 258,048
BatchNorm2d-261 [-1, 192, 17, 17] 384
ReLU-262 [-1, 192, 17, 17] 0
BasicConv2d-263 [-1, 192, 17, 17] 0
Conv2d-264 [-1, 192, 17, 17] 147,456
BatchNorm2d-265 [-1, 192, 17, 17] 384
ReLU-266 [-1, 192, 17, 17] 0
BasicConv2d-267 [-1, 192, 17, 17] 0
Conv2d-268 [-1, 192, 17, 17] 258,048
BatchNorm2d-269 [-1, 192, 17, 17] 384
ReLU-270 [-1, 192, 17, 17] 0
BasicConv2d-271 [-1, 192, 17, 17] 0
Conv2d-272 [-1, 192, 17, 17] 258,048
BatchNorm2d-273 [-1, 192, 17, 17] 384
ReLU-274 [-1, 192, 17, 17] 0
BasicConv2d-275 [-1, 192, 17, 17] 0
Conv2d-276 [-1, 192, 17, 17] 258,048
BatchNorm2d-277 [-1, 192, 17, 17] 384
ReLU-278 [-1, 192, 17, 17] 0
BasicConv2d-279 [-1, 192, 17, 17] 0
Conv2d-280 [-1, 192, 17, 17] 258,048
BatchNorm2d-281 [-1, 192, 17, 17] 384
ReLU-282 [-1, 192, 17, 17] 0
BasicConv2d-283 [-1, 192, 17, 17] 0
Conv2d-284 [-1, 192, 17, 17] 147,456
BatchNorm2d-285 [-1, 192, 17, 17] 384
ReLU-286 [-1, 192, 17, 17] 0
BasicConv2d-287 [-1, 192, 17, 17] 0
InceptionB-288 [-1, 768, 17, 17] 0
Conv2d-289 [-1, 192, 17, 17] 147,456
BatchNorm2d-290 [-1, 192, 17, 17] 384
ReLU-291 [-1, 192, 17, 17] 0
BasicConv2d-292 [-1, 192, 17, 17] 0
Conv2d-293 [-1, 320, 8, 8] 552,960
BatchNorm2d-294 [-1, 320, 8, 8] 640
ReLU-295 [-1, 320, 8, 8] 0
BasicConv2d-296 [-1, 320, 8, 8] 0
Conv2d-297 [-1, 192, 17, 17] 147,456
BatchNorm2d-298 [-1, 192, 17, 17] 384
ReLU-299 [-1, 192, 17, 17] 0
BasicConv2d-300 [-1, 192, 17, 17] 0
Conv2d-301 [-1, 192, 17, 17] 258,048
BatchNorm2d-302 [-1, 192, 17, 17] 384
ReLU-303 [-1, 192, 17, 17] 0
BasicConv2d-304 [-1, 192, 17, 17] 0
Conv2d-305 [-1, 192, 17, 17] 258,048
BatchNorm2d-306 [-1, 192, 17, 17] 384
ReLU-307 [-1, 192, 17, 17] 0
BasicConv2d-308 [-1, 192, 17, 17] 0
Conv2d-309 [-1, 192, 8, 8] 331,776
BatchNorm2d-310 [-1, 192, 8, 8] 384
ReLU-311 [-1, 192, 8, 8] 0
BasicConv2d-312 [-1, 192, 8, 8] 0
ReductionB-313 [-1, 1280, 8, 8] 0
Conv2d-314 [-1, 320, 8, 8] 409,600
BatchNorm2d-315 [-1, 320, 8, 8] 640
ReLU-316 [-1, 320, 8, 8] 0
BasicConv2d-317 [-1, 320, 8, 8] 0
Conv2d-318 [-1, 384, 8, 8] 491,520
BatchNorm2d-319 [-1, 384, 8, 8] 768
ReLU-320 [-1, 384, 8, 8] 0
BasicConv2d-321 [-1, 384, 8, 8] 0
Conv2d-322 [-1, 384, 8, 8] 442,368
BatchNorm2d-323 [-1, 384, 8, 8] 768
ReLU-324 [-1, 384, 8, 8] 0
BasicConv2d-325 [-1, 384, 8, 8] 0
Conv2d-326 [-1, 384, 8, 8] 442,368
BatchNorm2d-327 [-1, 384, 8, 8] 768
ReLU-328 [-1, 384, 8, 8] 0
BasicConv2d-329 [-1, 384, 8, 8] 0
Conv2d-330 [-1, 448, 8, 8] 573,440
BatchNorm2d-331 [-1, 448, 8, 8] 896
ReLU-332 [-1, 448, 8, 8] 0
BasicConv2d-333 [-1, 448, 8, 8] 0
Conv2d-334 [-1, 384, 8, 8] 1,548,288
BatchNorm2d-335 [-1, 384, 8, 8] 768
ReLU-336 [-1, 384, 8, 8] 0
BasicConv2d-337 [-1, 384, 8, 8] 0
Conv2d-338 [-1, 384, 8, 8] 442,368
BatchNorm2d-339 [-1, 384, 8, 8] 768
ReLU-340 [-1, 384, 8, 8] 0
BasicConv2d-341 [-1, 384, 8, 8] 0
Conv2d-342 [-1, 384, 8, 8] 442,368
BatchNorm2d-343 [-1, 384, 8, 8] 768
ReLU-344 [-1, 384, 8, 8] 0
BasicConv2d-345 [-1, 384, 8, 8] 0
Conv2d-346 [-1, 192, 8, 8] 245,760
BatchNorm2d-347 [-1, 192, 8, 8] 384
ReLU-348 [-1, 192, 8, 8] 0
BasicConv2d-349 [-1, 192, 8, 8] 0
InceptionC-350 [-1, 2048, 8, 8] 0
Conv2d-351 [-1, 320, 8, 8] 655,360
BatchNorm2d-352 [-1, 320, 8, 8] 640
ReLU-353 [-1, 320, 8, 8] 0
BasicConv2d-354 [-1, 320, 8, 8] 0
Conv2d-355 [-1, 384, 8, 8] 786,432
BatchNorm2d-356 [-1, 384, 8, 8] 768
ReLU-357 [-1, 384, 8, 8] 0
BasicConv2d-358 [-1, 384, 8, 8] 0
Conv2d-359 [-1, 384, 8, 8] 442,368
BatchNorm2d-360 [-1, 384, 8, 8] 768
ReLU-361 [-1, 384, 8, 8] 0
BasicConv2d-362 [-1, 384, 8, 8] 0
Conv2d-363 [-1, 384, 8, 8] 442,368
BatchNorm2d-364 [-1, 384, 8, 8] 768
ReLU-365 [-1, 384, 8, 8] 0
BasicConv2d-366 [-1, 384, 8, 8] 0
Conv2d-367 [-1, 448, 8, 8] 917,504
BatchNorm2d-368 [-1, 448, 8, 8] 896
ReLU-369 [-1, 448, 8, 8] 0
BasicConv2d-370 [-1, 448, 8, 8] 0
Conv2d-371 [-1, 384, 8, 8] 1,548,288
BatchNorm2d-372 [-1, 384, 8, 8] 768
ReLU-373 [-1, 384, 8, 8] 0
BasicConv2d-374 [-1, 384, 8, 8] 0
Conv2d-375 [-1, 384, 8, 8] 442,368
BatchNorm2d-376 [-1, 384, 8, 8] 768
ReLU-377 [-1, 384, 8, 8] 0
BasicConv2d-378 [-1, 384, 8, 8] 0
Conv2d-379 [-1, 384, 8, 8] 442,368
BatchNorm2d-380 [-1, 384, 8, 8] 768
ReLU-381 [-1, 384, 8, 8] 0
BasicConv2d-382 [-1, 384, 8, 8] 0
Conv2d-383 [-1, 192, 8, 8] 393,216
BatchNorm2d-384 [-1, 192, 8, 8] 384
ReLU-385 [-1, 192, 8, 8] 0
BasicConv2d-386 [-1, 192, 8, 8] 0
InceptionC-387 [-1, 2048, 8, 8] 0
Linear-388 [-1, 4] 8,196
================================================================
Total params: 21,793,764
Trainable params: 21,793,764
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.02
Forward/backward pass size (MB): 292.53
Params size (MB): 83.14
Estimated Total Size (MB): 376.69
----------------------------------------------------------------三、 训练模型
1. 编写训练函数
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
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，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_loss2. 编写测试函数
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
#测试函数
def test(dataloader,model,loss_fn):
size=len(dataloader.dataset) #测试集的大小
num_batches=len(dataloader) #批次数目
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_loss3. 正式训练
import copy
opt=torch.optim.Adam(model.parameters(),lr=1e-4) #创建优化器，并设置学习率
loss_fn=nn.CrossEntropyLoss() #创建损失函数
epochs=50
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,opt)
model.eval()
epoch_test_acc,epoch_test_loss=test(test_dl,model,loss_fn)
#保存最佳模型到J9_model
if epoch_test_acc>best_acc:
best_acc=epoch_test_acc
J9_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)
#获取当前学习率
lr=opt.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))
#保存最佳模型到文件中
PATH=r'D:\THE MNIST DATABASE\J-series\J9_model.pth'
torch.save(model.state_dict(),PATH)运行结果：
Epoch: 1,Train_acc:77.2%,Train_loss:0.634,Test_acc:89.3%,Test_loss:0.443,Lr:1.00E-04
Epoch: 2,Train_acc:78.3%,Train_loss:0.596,Test_acc:88.9%,Test_loss:0.346,Lr:1.00E-04
Epoch: 3,Train_acc:82.3%,Train_loss:0.477,Test_acc:68.9%,Test_loss:1.772,Lr:1.00E-04
Epoch: 4,Train_acc:83.1%,Train_loss:0.510,Test_acc:88.0%,Test_loss:0.358,Lr:1.00E-04
Epoch: 5,Train_acc:84.8%,Train_loss:0.443,Test_acc:91.1%,Test_loss:0.213,Lr:1.00E-04
Epoch: 6,Train_acc:83.9%,Train_loss:0.432,Test_acc:88.4%,Test_loss:0.301,Lr:1.00E-04
Epoch: 7,Train_acc:87.8%,Train_loss:0.368,Test_acc:88.4%,Test_loss:0.300,Lr:1.00E-04
Epoch: 8,Train_acc:87.9%,Train_loss:0.371,Test_acc:90.2%,Test_loss:0.358,Lr:1.00E-04
Epoch: 9,Train_acc:88.6%,Train_loss:0.361,Test_acc:92.9%,Test_loss:0.183,Lr:1.00E-04
Epoch:10,Train_acc:88.9%,Train_loss:0.310,Test_acc:91.6%,Test_loss:0.277,Lr:1.00E-04
Epoch:11,Train_acc:89.8%,Train_loss:0.306,Test_acc:94.2%,Test_loss:0.174,Lr:1.00E-04
Epoch:12,Train_acc:88.0%,Train_loss:0.357,Test_acc:92.0%,Test_loss:0.301,Lr:1.00E-04
Epoch:13,Train_acc:92.1%,Train_loss:0.246,Test_acc:92.9%,Test_loss:0.266,Lr:1.00E-04
Epoch:14,Train_acc:91.2%,Train_loss:0.269,Test_acc:94.2%,Test_loss:0.163,Lr:1.00E-04
Epoch:15,Train_acc:90.8%,Train_loss:0.294,Test_acc:87.1%,Test_loss:0.331,Lr:1.00E-04
Epoch:16,Train_acc:91.4%,Train_loss:0.251,Test_acc:93.8%,Test_loss:0.177,Lr:1.00E-04
Epoch:17,Train_acc:92.9%,Train_loss:0.214,Test_acc:90.7%,Test_loss:0.276,Lr:1.00E-04
Epoch:18,Train_acc:88.9%,Train_loss:0.315,Test_acc:92.9%,Test_loss:0.215,Lr:1.00E-04
Epoch:19,Train_acc:92.2%,Train_loss:0.231,Test_acc:91.6%,Test_loss:0.251,Lr:1.00E-04
Epoch:20,Train_acc:92.9%,Train_loss:0.228,Test_acc:91.1%,Test_loss:0.229,Lr:1.00E-04
Epoch:21,Train_acc:93.0%,Train_loss:0.211,Test_acc:90.7%,Test_loss:0.245,Lr:1.00E-04
Epoch:22,Train_acc:93.2%,Train_loss:0.235,Test_acc:94.2%,Test_loss:0.150,Lr:1.00E-04
Epoch:23,Train_acc:91.6%,Train_loss:0.259,Test_acc:83.1%,Test_loss:0.375,Lr:1.00E-04
Epoch:24,Train_acc:93.2%,Train_loss:0.209,Test_acc:93.3%,Test_loss:0.219,Lr:1.00E-04
Epoch:25,Train_acc:93.2%,Train_loss:0.183,Test_acc:94.2%,Test_loss:0.181,Lr:1.00E-04
Epoch:26,Train_acc:91.4%,Train_loss:0.242,Test_acc:92.9%,Test_loss:0.249,Lr:1.00E-04
Epoch:27,Train_acc:92.0%,Train_loss:0.220,Test_acc:93.3%,Test_loss:0.234,Lr:1.00E-04
Epoch:28,Train_acc:94.0%,Train_loss:0.195,Test_acc:94.7%,Test_loss:0.141,Lr:1.00E-04
Epoch:29,Train_acc:93.1%,Train_loss:0.189,Test_acc:95.1%,Test_loss:0.191,Lr:1.00E-04
Epoch:30,Train_acc:93.2%,Train_loss:0.226,Test_acc:95.6%,Test_loss:0.204,Lr:1.00E-04
Epoch:31,Train_acc:94.3%,Train_loss:0.147,Test_acc:93.8%,Test_loss:0.219,Lr:1.00E-04
Epoch:32,Train_acc:95.4%,Train_loss:0.150,Test_acc:93.3%,Test_loss:0.225,Lr:1.00E-04
Epoch:33,Train_acc:94.9%,Train_loss:0.165,Test_acc:95.6%,Test_loss:0.169,Lr:1.00E-04
Epoch:34,Train_acc:95.4%,Train_loss:0.150,Test_acc:94.7%,Test_loss:0.193,Lr:1.00E-04
Epoch:35,Train_acc:95.7%,Train_loss:0.100,Test_acc:93.8%,Test_loss:0.145,Lr:1.00E-04
Epoch:36,Train_acc:94.4%,Train_loss:0.198,Test_acc:92.9%,Test_loss:0.167,Lr:1.00E-04
Epoch:37,Train_acc:95.3%,Train_loss:0.163,Test_acc:94.7%,Test_loss:0.110,Lr:1.00E-04
Epoch:38,Train_acc:96.1%,Train_loss:0.120,Test_acc:93.3%,Test_loss:0.177,Lr:1.00E-04
Epoch:39,Train_acc:94.6%,Train_loss:0.197,Test_acc:94.2%,Test_loss:0.196,Lr:1.00E-04
Epoch:40,Train_acc:95.4%,Train_loss:0.132,Test_acc:96.0%,Test_loss:0.117,Lr:1.00E-04
Epoch:41,Train_acc:96.6%,Train_loss:0.115,Test_acc:96.9%,Test_loss:0.116,Lr:1.00E-04
Epoch:42,Train_acc:96.1%,Train_loss:0.113,Test_acc:95.6%,Test_loss:0.119,Lr:1.00E-04
Epoch:43,Train_acc:97.1%,Train_loss:0.103,Test_acc:93.3%,Test_loss:0.218,Lr:1.00E-04
Epoch:44,Train_acc:94.9%,Train_loss:0.168,Test_acc:89.3%,Test_loss:0.251,Lr:1.00E-04
Epoch:45,Train_acc:97.3%,Train_loss:0.094,Test_acc:93.3%,Test_loss:0.180,Lr:1.00E-04
Epoch:46,Train_acc:97.4%,Train_loss:0.086,Test_acc:94.7%,Test_loss:0.210,Lr:1.00E-04
Epoch:47,Train_acc:95.3%,Train_loss:0.125,Test_acc:95.1%,Test_loss:0.200,Lr:1.00E-04
Epoch:48,Train_acc:95.9%,Train_loss:0.131,Test_acc:94.7%,Test_loss:0.159,Lr:1.00E-04
Epoch:49,Train_acc:95.0%,Train_loss:0.147,Test_acc:93.8%,Test_loss:0.218,Lr:1.00E-04
Epoch:50,Train_acc:97.6%,Train_loss:0.076,Test_acc:95.6%,Test_loss:0.172,Lr:1.00E-04四、 结果可视化
1. Loss与Accuracy图
import matplotlib.pyplot as plt
#隐藏警告
import warnings
warnings.filterwarnings("ignore") #忽略警告信息
plt.rcParams['font.sans-serif']=['SimHei'] #正常显示中文标签
plt.rcParams['axes.unicode_minus']=False #正常显示负号
plt.rcParams['figure.dpi']=300 #分辨率
epochs_range=range(epochs)
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()运行结果：
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}')预测图片：
#预测训练集中的某张照片
predict_one_image(image_path=r'D:\THE MNIST DATABASE\weather_photos\shine\shine15.jpg',
model=model,transform=train_transforms,classes=classes)运行结果：
预测结果是：shine
五、心得体会
在本周的项目训练中，手动搭建了InceptionV3模型，加深了对该模型的理解。