- 🍨 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|接辅导、项目定制
目录
一、课题背景和开发环境
📌第J9周：Inception v3算法实战与解析📌
- 语言：Python3、Pytorch
📌本周任务：📌
- 了解并学习InceptionV3相对于InceptionV1改进了哪些地方（重点）
- 使用Inception v3完成天气识别案例
二、理论基础
Inception v3 论文
Rethinking the Inception Architecture for Computer Vision.pdf
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（如上图所示），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：
三、搭建网络模型
0.BasicConv2d
class BasicConv2d(nn.Module):
def __init__(self, in_channel, out_channel, **kwargs):
super(BasicConv2d, self).__init__()
self.conv = nn.Conv2d(in_channel, out_channel, bias=False, **kwargs)
self.norm = nn.BatchNorm2d(out_channel, eps=0.001)
self.relu = nn.ReLU(inplace=True)
def forward(self, x):
x = self.conv(x)
x = self.norm(x)
x = self.relu(x)
return x
1.Inception-A
class InceptionA(nn.Module):
def __init__(self, in_channel, pool_feature):
super(InceptionA, self).__init__()
# 1x1 conv
self.conv1 = BasicConv2d(in_channel, 64, kernel_size=1)
# 1x1 conv -> 5x5 conv
self.conv2 = nn.Sequential(
BasicConv2d(in_channel, 48, kernel_size=1),
BasicConv2d(48, 64, kernel_size=5, padding=2)
)
# 1x1 conv -> [3x3 conv]x2
self.conv3 = nn.Sequential(
BasicConv2d(in_channel, 64, kernel_size=1),
BasicConv2d(64, 96, kernel_size=3, padding=1),
BasicConv2d(96, 96, kernel_size=3, padding=1)
)
# 3x3 avgpool -> 1x1 conv
self.conv4 = nn.Sequential(
nn.AvgPool2d(kernel_size=3, stride=1, padding=1),
BasicConv2d(in_channel, pool_feature, kernel_size=1)
)
def forward(self, x):
x1 = self.conv1(x)
x2 = self.conv2(x)
x3 = self.conv3(x)
x4 = self.conv4(x)
x = torch.cat([x1,x2,x3,x4], 1)
return x
2.Inception-B
class InceptionB(nn.Module):
def __init__(self, in_channel, filter7x7):
super(InceptionB, self).__init__()
# 1x1 conv
self.conv1 = BasicConv2d(in_channel, 192, kernel_size=1)
# 1x1 conv -> 1x7 conv -> 7x1 conv
self.conv2 = nn.Sequential(
BasicConv2d(in_channel, filter7x7, kernel_size=1),
BasicConv2d(filter7x7, filter7x7, kernel_size=(1,7), padding=(0,3)),
BasicConv2d(filter7x7, 192, kernel_size=(7,1), padding=(3,0))
)
# 1x1 conv -> [7x1 conv -> 1x7 conv]x2
self.conv3 = nn.Sequential(
BasicConv2d(in_channel, filter7x764, kernel_size=1),
BasicConv2d(filter7x7, filter7x7, kernel_size=(7,1), padding=(3,0)),
BasicConv2d(filter7x7, filter7x7, kernel_size=(1,7), padding=(0,3)),
BasicConv2d(filter7x7, filter7x7, kernel_size=(7,1), padding=(3,0)),
BasicConv2d(filter7x7, 192, kernel_size=(1,7), padding=(0,3))
)
# 3x3 avgpool -> 1x1 conv
self.conv4 = nn.Sequential(
nn.AvgPool2d(kernel_size=3, stride=1, padding=1),
BasicConv2d(in_channel, 192, kernel_size=1)
)
def forward(self, x):
x1 = self.conv1(x)
x2 = self.conv2(x)
x3 = self.conv3(x)
x4 = self.conv4(x)
x = torch.cat([x1,x2,x3,x4], 1)
return x
3.Inception-C
class InceptionC(nn.Module):
def __init__(self, in_channel):
super(InceptionC, self).__init__()
# 1x1 conv
self.conv1 = BasicConv2d(in_channel, 320, kernel_size=1)
# 1x1 conv -> [1x3 conv, 3x1 conv]
self.conv2 = BasicConv2d(in_channel, 384, kernel_size=1)
self.conv2a = BasicConv2d(384, 384, kernel_size=(1,3), padding=(0,1))
self.conv2b = BasicConv2d(384, 384, kernel_size=(3,1), padding=(1,0))
# 1x1 conv -> [3x3 conv]x2
self.conv3 = nn.Sequential(
BasicConv2d(in_channel, 448, kernel_size=1),
BasicConv2d(448, 384, kernel_size=3, padding=1)
)
self.conv3a = BasicConv2d(384, 384, kernel_size=(1,3), padding=(0,1))
self.conv3b = BasicConv2d(384, 384, kernel_size=(3,1), padding=(1,0))
# 3x3 avgpool -> 1x1 conv
self.conv4 = nn.Sequential(
nn.AvgPool2d(kernel_size=3, stride=1, padding=1),
BasicConv2d(in_channel, 192, kernel_size=1)
)
def forward(self, x):
x1 = self.conv1(x)
x2 = self.conv2(x)
x2 = torch.cat([self.conv2a(x2),self.conv2b(x2)], 1)
x3 = self.conv3(x)
x3 = torch.cat([self.conv2a(x3),self.conv2b(x3)], 1)
x4 = self.conv4(x)
x = torch.cat([x1,x2,x3,x4], 1)
return x
4.Reduction-A
class ReductionA(nn.Module):
def __init__(self, in_channel):
super(ReductionA, self).__init__()
# 3x3 conv
self.conv1 = BasicConv2d(in_channel, 384, kernel_size=3, stride=2)
# 1x1 conv -> [3x3 conv]x2
self.conv2 = nn.Sequential(
BasicConv2d(in_channel, 64, kernel_size=1),
BasicConv2d(64, 96, kernel_size=3, padding=1),
BasicConv2d(96, 96, kernel_size=3, stride=2)
)
# 3x3 maxpool
self.conv3 = nn.MaxPool2d(kernel_size=3, stride=2)
def forward(self, x):
x1 = self.conv1(x)
x2 = self.conv2(x)
x3 = self.conv3(x)
x = torch.cat([x1,x2,x3], 1)
return x
5.Reduction-B
class ReductionB(nn.Module):
def __init__(self, in_channel):
super(ReductionB, self).__init__()
# 1x1 conv -> 3x3 conv
self.conv1 = nn.Sequential(
BasicConv2d(in_channel, 192, kernel_size=1),
BasicConv2d(192, 320, kernel_size=3, stride=2)
)
# 1x1 conv -> 1x7 conv -> 7x1 conv -> 3x3 conv
self.conv2 = nn.Sequential(
BasicConv2d(in_channel, 192, kernel_size=1),
BasicConv2d(192, 192, kernel_size=(1,7), padding=(0,3)),
BasicConv2d(192, 192, kernel_size=(7,1), padding=(3,0)),
BasicConv2d(192, 192, kernel_size=3, stride=2)
)
# 3x3 maxpool
self.conv3 = nn.MaxPool2d(kernel_size=3, stride=2)
def forward(self, x):
x1 = self.conv1(x)
x2 = self.conv2(x)
x3 = self.conv3(x)
x = torch.cat([x1,x2,x3], 1)
return x
6.辅助分支
class InceptionAux(nn.Module):
def __init__(self, in_channel, num_classes):
super(InceptionAux, self).__init__()
self.pool0 = nn.AvgPool2d(kernel_size=5, stride=3)
self.conv1 = BasicConv2d(in_channel, 128, kernel_size=1)
self.conv2 = BasicConv2d(128, 768, kernel_size=5)
self.conv2.stddev = 0.01
self.fc = nn.Linear(768, num_classes)
self.fc.stddev = 0.001
def forward(self, x):
x = self.pool0(x)
x = self.conv1(x)
x = self.conv2(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
return x
7.模型搭建
''' InceptionV3 '''
class InceptionV3(nn.Module):
def __init__(self,
input_shape=[3, 224, 224],
classes=1000, # 用于分类图像的可选类数
aux_logits=False,
transform_input=False):
super(InceptionV3, self).__init__()
self.aux_logits = aux_logits
self.transform_input = transform_input
self.stem = nn.Sequential(
BasicConv2d(3, 32, kernel_size=3, stride=2),
BasicConv2d(32, 32, kernel_size=3),
BasicConv2d(32, 64, kernel_size=3, padding=1),
nn.MaxPool2d(kernel_size=3, stride=2),
BasicConv2d(64, 80, kernel_size=1),
BasicConv2d(80, 192, kernel_size=3),
nn.MaxPool2d(kernel_size=3, stride=2)
)
self.block1 = nn.Sequential(
InceptionA(192, pool_feature=32),
InceptionA(256, pool_feature=64),
InceptionA(288, pool_feature=64)
)
self.block2 = nn.Sequential(
ReductionA(288),
InceptionB(768, filter7x7=128),
InceptionB(768, filter7x7=160),
InceptionB(768, filter7x7=160),
InceptionB(768, filter7x7=192)
)
if self.aux_logits:
self.AuxLogits = InceptionAux(768, classes)
self.block3 = nn.Sequential(
ReductionB(768),
InceptionC(1280),
InceptionC(2048)
)
self.fc = nn.Linear(2048, 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
x = self.stem(x)
x = self.block1(x)
x = self.block2(x)
if self.training and self.aux_logits:
aux = self.AuxLogits(x)
x = self.block3(x)
x = nn.AvgPool2d(kernel_size=8)(x)
x = F.dropout(x, training=self.training)
x = x.view(x.size(0), -1)
x = self.fc(x)
if self.training and self.aux_logits:
return x, aux
return x
8.查看模型
----------------------------------------------------------------
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
MaxPool2d-13 [-1, 64, 73, 73] 0
Conv2d-14 [-1, 80, 73, 73] 5,120
BatchNorm2d-15 [-1, 80, 73, 73] 160
ReLU-16 [-1, 80, 73, 73] 0
BasicConv2d-17 [-1, 80, 73, 73] 0
Conv2d-18 [-1, 192, 71, 71] 138,240
BatchNorm2d-19 [-1, 192, 71, 71] 384
ReLU-20 [-1, 192, 71, 71] 0
BasicConv2d-21 [-1, 192, 71, 71] 0
MaxPool2d-22 [-1, 192, 35, 35] 0
Conv2d-23 [-1, 64, 35, 35] 12,288
BatchNorm2d-24 [-1, 64, 35, 35] 128
ReLU-25 [-1, 64, 35, 35] 0
BasicConv2d-26 [-1, 64, 35, 35] 0
Conv2d-27 [-1, 48, 35, 35] 9,216
BatchNorm2d-28 [-1, 48, 35, 35] 96
ReLU-29 [-1, 48, 35, 35] 0
BasicConv2d-30 [-1, 48, 35, 35] 0
Conv2d-31 [-1, 64, 35, 35] 76,800
BatchNorm2d-32 [-1, 64, 35, 35] 128
ReLU-33 [-1, 64, 35, 35] 0
BasicConv2d-34 [-1, 64, 35, 35] 0
Conv2d-35 [-1, 64, 35, 35] 12,288
BatchNorm2d-36 [-1, 64, 35, 35] 128
ReLU-37 [-1, 64, 35, 35] 0
BasicConv2d-38 [-1, 64, 35, 35] 0
Conv2d-39 [-1, 96, 35, 35] 55,296
BatchNorm2d-40 [-1, 96, 35, 35] 192
ReLU-41 [-1, 96, 35, 35] 0
BasicConv2d-42 [-1, 96, 35, 35] 0
Conv2d-43 [-1, 96, 35, 35] 82,944
BatchNorm2d-44 [-1, 96, 35, 35] 192
ReLU-45 [-1, 96, 35, 35] 0
BasicConv2d-46 [-1, 96, 35, 35] 0
AvgPool2d-47 [-1, 192, 35, 35] 0
Conv2d-48 [-1, 32, 35, 35] 6,144
BatchNorm2d-49 [-1, 32, 35, 35] 64
ReLU-50 [-1, 32, 35, 35] 0
BasicConv2d-51 [-1, 32, 35, 35] 0
InceptionA-52 [-1, 256, 35, 35] 0
Conv2d-53 [-1, 64, 35, 35] 16,384
BatchNorm2d-54 [-1, 64, 35, 35] 128
ReLU-55 [-1, 64, 35, 35] 0
BasicConv2d-56 [-1, 64, 35, 35] 0
Conv2d-57 [-1, 48, 35, 35] 12,288
BatchNorm2d-58 [-1, 48, 35, 35] 96
ReLU-59 [-1, 48, 35, 35] 0
BasicConv2d-60 [-1, 48, 35, 35] 0
Conv2d-61 [-1, 64, 35, 35] 76,800
BatchNorm2d-62 [-1, 64, 35, 35] 128
ReLU-63 [-1, 64, 35, 35] 0
BasicConv2d-64 [-1, 64, 35, 35] 0
Conv2d-65 [-1, 64, 35, 35] 16,384
BatchNorm2d-66 [-1, 64, 35, 35] 128
ReLU-67 [-1, 64, 35, 35] 0
BasicConv2d-68 [-1, 64, 35, 35] 0
Conv2d-69 [-1, 96, 35, 35] 55,296
BatchNorm2d-70 [-1, 96, 35, 35] 192
ReLU-71 [-1, 96, 35, 35] 0
BasicConv2d-72 [-1, 96, 35, 35] 0
Conv2d-73 [-1, 96, 35, 35] 82,944
BatchNorm2d-74 [-1, 96, 35, 35] 192
ReLU-75 [-1, 96, 35, 35] 0
BasicConv2d-76 [-1, 96, 35, 35] 0
AvgPool2d-77 [-1, 256, 35, 35] 0
Conv2d-78 [-1, 64, 35, 35] 16,384
BatchNorm2d-79 [-1, 64, 35, 35] 128
ReLU-80 [-1, 64, 35, 35] 0
BasicConv2d-81 [-1, 64, 35, 35] 0
InceptionA-82 [-1, 288, 35, 35] 0
Conv2d-83 [-1, 64, 35, 35] 18,432
BatchNorm2d-84 [-1, 64, 35, 35] 128
ReLU-85 [-1, 64, 35, 35] 0
BasicConv2d-86 [-1, 64, 35, 35] 0
Conv2d-87 [-1, 48, 35, 35] 13,824
BatchNorm2d-88 [-1, 48, 35, 35] 96
ReLU-89 [-1, 48, 35, 35] 0
BasicConv2d-90 [-1, 48, 35, 35] 0
Conv2d-91 [-1, 64, 35, 35] 76,800
BatchNorm2d-92 [-1, 64, 35, 35] 128
ReLU-93 [-1, 64, 35, 35] 0
BasicConv2d-94 [-1, 64, 35, 35] 0
Conv2d-95 [-1, 64, 35, 35] 18,432
BatchNorm2d-96 [-1, 64, 35, 35] 128
ReLU-97 [-1, 64, 35, 35] 0
BasicConv2d-98 [-1, 64, 35, 35] 0
Conv2d-99 [-1, 96, 35, 35] 55,296
BatchNorm2d-100 [-1, 96, 35, 35] 192
ReLU-101 [-1, 96, 35, 35] 0
BasicConv2d-102 [-1, 96, 35, 35] 0
Conv2d-103 [-1, 96, 35, 35] 82,944
BatchNorm2d-104 [-1, 96, 35, 35] 192
ReLU-105 [-1, 96, 35, 35] 0
BasicConv2d-106 [-1, 96, 35, 35] 0
AvgPool2d-107 [-1, 288, 35, 35] 0
Conv2d-108 [-1, 64, 35, 35] 18,432
BatchNorm2d-109 [-1, 64, 35, 35] 128
ReLU-110 [-1, 64, 35, 35] 0
BasicConv2d-111 [-1, 64, 35, 35] 0
InceptionA-112 [-1, 288, 35, 35] 0
Conv2d-113 [-1, 384, 17, 17] 995,328
BatchNorm2d-114 [-1, 384, 17, 17] 768
ReLU-115 [-1, 384, 17, 17] 0
BasicConv2d-116 [-1, 384, 17, 17] 0
Conv2d-117 [-1, 64, 35, 35] 18,432
BatchNorm2d-118 [-1, 64, 35, 35] 128
ReLU-119 [-1, 64, 35, 35] 0
BasicConv2d-120 [-1, 64, 35, 35] 0
Conv2d-121 [-1, 96, 35, 35] 55,296
BatchNorm2d-122 [-1, 96, 35, 35] 192
ReLU-123 [-1, 96, 35, 35] 0
BasicConv2d-124 [-1, 96, 35, 35] 0
Conv2d-125 [-1, 96, 17, 17] 82,944
BatchNorm2d-126 [-1, 96, 17, 17] 192
ReLU-127 [-1, 96, 17, 17] 0
BasicConv2d-128 [-1, 96, 17, 17] 0
MaxPool2d-129 [-1, 288, 17, 17] 0
ReductionA-130 [-1, 768, 17, 17] 0
Conv2d-131 [-1, 192, 17, 17] 147,456
BatchNorm2d-132 [-1, 192, 17, 17] 384
ReLU-133 [-1, 192, 17, 17] 0
BasicConv2d-134 [-1, 192, 17, 17] 0
Conv2d-135 [-1, 128, 17, 17] 98,304
BatchNorm2d-136 [-1, 128, 17, 17] 256
ReLU-137 [-1, 128, 17, 17] 0
BasicConv2d-138 [-1, 128, 17, 17] 0
Conv2d-139 [-1, 128, 17, 17] 114,688
BatchNorm2d-140 [-1, 128, 17, 17] 256
ReLU-141 [-1, 128, 17, 17] 0
BasicConv2d-142 [-1, 128, 17, 17] 0
Conv2d-143 [-1, 192, 17, 17] 172,032
BatchNorm2d-144 [-1, 192, 17, 17] 384
ReLU-145 [-1, 192, 17, 17] 0
BasicConv2d-146 [-1, 192, 17, 17] 0
Conv2d-147 [-1, 128, 17, 17] 98,304
BatchNorm2d-148 [-1, 128, 17, 17] 256
ReLU-149 [-1, 128, 17, 17] 0
BasicConv2d-150 [-1, 128, 17, 17] 0
Conv2d-151 [-1, 128, 17, 17] 114,688
BatchNorm2d-152 [-1, 128, 17, 17] 256
ReLU-153 [-1, 128, 17, 17] 0
BasicConv2d-154 [-1, 128, 17, 17] 0
Conv2d-155 [-1, 128, 17, 17] 114,688
BatchNorm2d-156 [-1, 128, 17, 17] 256
ReLU-157 [-1, 128, 17, 17] 0
BasicConv2d-158 [-1, 128, 17, 17] 0
Conv2d-159 [-1, 128, 17, 17] 114,688
BatchNorm2d-160 [-1, 128, 17, 17] 256
ReLU-161 [-1, 128, 17, 17] 0
BasicConv2d-162 [-1, 128, 17, 17] 0
Conv2d-163 [-1, 192, 17, 17] 172,032
BatchNorm2d-164 [-1, 192, 17, 17] 384
ReLU-165 [-1, 192, 17, 17] 0
BasicConv2d-166 [-1, 192, 17, 17] 0
AvgPool2d-167 [-1, 768, 17, 17] 0
Conv2d-168 [-1, 192, 17, 17] 147,456
BatchNorm2d-169 [-1, 192, 17, 17] 384
ReLU-170 [-1, 192, 17, 17] 0
BasicConv2d-171 [-1, 192, 17, 17] 0
InceptionB-172 [-1, 768, 17, 17] 0
Conv2d-173 [-1, 192, 17, 17] 147,456
BatchNorm2d-174 [-1, 192, 17, 17] 384
ReLU-175 [-1, 192, 17, 17] 0
BasicConv2d-176 [-1, 192, 17, 17] 0
Conv2d-177 [-1, 160, 17, 17] 122,880
BatchNorm2d-178 [-1, 160, 17, 17] 320
ReLU-179 [-1, 160, 17, 17] 0
BasicConv2d-180 [-1, 160, 17, 17] 0
Conv2d-181 [-1, 160, 17, 17] 179,200
BatchNorm2d-182 [-1, 160, 17, 17] 320
ReLU-183 [-1, 160, 17, 17] 0
BasicConv2d-184 [-1, 160, 17, 17] 0
Conv2d-185 [-1, 192, 17, 17] 215,040
BatchNorm2d-186 [-1, 192, 17, 17] 384
ReLU-187 [-1, 192, 17, 17] 0
BasicConv2d-188 [-1, 192, 17, 17] 0
Conv2d-189 [-1, 160, 17, 17] 122,880
BatchNorm2d-190 [-1, 160, 17, 17] 320
ReLU-191 [-1, 160, 17, 17] 0
BasicConv2d-192 [-1, 160, 17, 17] 0
Conv2d-193 [-1, 160, 17, 17] 179,200
BatchNorm2d-194 [-1, 160, 17, 17] 320
ReLU-195 [-1, 160, 17, 17] 0
BasicConv2d-196 [-1, 160, 17, 17] 0
Conv2d-197 [-1, 160, 17, 17] 179,200
BatchNorm2d-198 [-1, 160, 17, 17] 320
ReLU-199 [-1, 160, 17, 17] 0
BasicConv2d-200 [-1, 160, 17, 17] 0
Conv2d-201 [-1, 160, 17, 17] 179,200
BatchNorm2d-202 [-1, 160, 17, 17] 320
ReLU-203 [-1, 160, 17, 17] 0
BasicConv2d-204 [-1, 160, 17, 17] 0
Conv2d-205 [-1, 192, 17, 17] 215,040
BatchNorm2d-206 [-1, 192, 17, 17] 384
ReLU-207 [-1, 192, 17, 17] 0
BasicConv2d-208 [-1, 192, 17, 17] 0
AvgPool2d-209 [-1, 768, 17, 17] 0
Conv2d-210 [-1, 192, 17, 17] 147,456
BatchNorm2d-211 [-1, 192, 17, 17] 384
ReLU-212 [-1, 192, 17, 17] 0
BasicConv2d-213 [-1, 192, 17, 17] 0
InceptionB-214 [-1, 768, 17, 17] 0
Conv2d-215 [-1, 192, 17, 17] 147,456
BatchNorm2d-216 [-1, 192, 17, 17] 384
ReLU-217 [-1, 192, 17, 17] 0
BasicConv2d-218 [-1, 192, 17, 17] 0
Conv2d-219 [-1, 160, 17, 17] 122,880
BatchNorm2d-220 [-1, 160, 17, 17] 320
ReLU-221 [-1, 160, 17, 17] 0
BasicConv2d-222 [-1, 160, 17, 17] 0
Conv2d-223 [-1, 160, 17, 17] 179,200
BatchNorm2d-224 [-1, 160, 17, 17] 320
ReLU-225 [-1, 160, 17, 17] 0
BasicConv2d-226 [-1, 160, 17, 17] 0
Conv2d-227 [-1, 192, 17, 17] 215,040
BatchNorm2d-228 [-1, 192, 17, 17] 384
ReLU-229 [-1, 192, 17, 17] 0
BasicConv2d-230 [-1, 192, 17, 17] 0
Conv2d-231 [-1, 160, 17, 17] 122,880
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, 160, 17, 17] 179,200
BatchNorm2d-240 [-1, 160, 17, 17] 320
ReLU-241 [-1, 160, 17, 17] 0
BasicConv2d-242 [-1, 160, 17, 17] 0
Conv2d-243 [-1, 160, 17, 17] 179,200
BatchNorm2d-244 [-1, 160, 17, 17] 320
ReLU-245 [-1, 160, 17, 17] 0
BasicConv2d-246 [-1, 160, 17, 17] 0
Conv2d-247 [-1, 192, 17, 17] 215,040
BatchNorm2d-248 [-1, 192, 17, 17] 384
ReLU-249 [-1, 192, 17, 17] 0
BasicConv2d-250 [-1, 192, 17, 17] 0
AvgPool2d-251 [-1, 768, 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
InceptionB-256 [-1, 768, 17, 17] 0
Conv2d-257 [-1, 192, 17, 17] 147,456
BatchNorm2d-258 [-1, 192, 17, 17] 384
ReLU-259 [-1, 192, 17, 17] 0
BasicConv2d-260 [-1, 192, 17, 17] 0
Conv2d-261 [-1, 192, 17, 17] 147,456
BatchNorm2d-262 [-1, 192, 17, 17] 384
ReLU-263 [-1, 192, 17, 17] 0
BasicConv2d-264 [-1, 192, 17, 17] 0
Conv2d-265 [-1, 192, 17, 17] 258,048
BatchNorm2d-266 [-1, 192, 17, 17] 384
ReLU-267 [-1, 192, 17, 17] 0
BasicConv2d-268 [-1, 192, 17, 17] 0
Conv2d-269 [-1, 192, 17, 17] 258,048
BatchNorm2d-270 [-1, 192, 17, 17] 384
ReLU-271 [-1, 192, 17, 17] 0
BasicConv2d-272 [-1, 192, 17, 17] 0
Conv2d-273 [-1, 192, 17, 17] 147,456
BatchNorm2d-274 [-1, 192, 17, 17] 384
ReLU-275 [-1, 192, 17, 17] 0
BasicConv2d-276 [-1, 192, 17, 17] 0
Conv2d-277 [-1, 192, 17, 17] 258,048
BatchNorm2d-278 [-1, 192, 17, 17] 384
ReLU-279 [-1, 192, 17, 17] 0
BasicConv2d-280 [-1, 192, 17, 17] 0
Conv2d-281 [-1, 192, 17, 17] 258,048
BatchNorm2d-282 [-1, 192, 17, 17] 384
ReLU-283 [-1, 192, 17, 17] 0
BasicConv2d-284 [-1, 192, 17, 17] 0
Conv2d-285 [-1, 192, 17, 17] 258,048
BatchNorm2d-286 [-1, 192, 17, 17] 384
ReLU-287 [-1, 192, 17, 17] 0
BasicConv2d-288 [-1, 192, 17, 17] 0
Conv2d-289 [-1, 192, 17, 17] 258,048
BatchNorm2d-290 [-1, 192, 17, 17] 384
ReLU-291 [-1, 192, 17, 17] 0
BasicConv2d-292 [-1, 192, 17, 17] 0
AvgPool2d-293 [-1, 768, 17, 17] 0
Conv2d-294 [-1, 192, 17, 17] 147,456
BatchNorm2d-295 [-1, 192, 17, 17] 384
ReLU-296 [-1, 192, 17, 17] 0
BasicConv2d-297 [-1, 192, 17, 17] 0
InceptionB-298 [-1, 768, 17, 17] 0
Conv2d-299 [-1, 192, 17, 17] 147,456
BatchNorm2d-300 [-1, 192, 17, 17] 384
ReLU-301 [-1, 192, 17, 17] 0
BasicConv2d-302 [-1, 192, 17, 17] 0
Conv2d-303 [-1, 320, 8, 8] 552,960
BatchNorm2d-304 [-1, 320, 8, 8] 640
ReLU-305 [-1, 320, 8, 8] 0
BasicConv2d-306 [-1, 320, 8, 8] 0
Conv2d-307 [-1, 192, 17, 17] 147,456
BatchNorm2d-308 [-1, 192, 17, 17] 384
ReLU-309 [-1, 192, 17, 17] 0
BasicConv2d-310 [-1, 192, 17, 17] 0
Conv2d-311 [-1, 192, 17, 17] 258,048
BatchNorm2d-312 [-1, 192, 17, 17] 384
ReLU-313 [-1, 192, 17, 17] 0
BasicConv2d-314 [-1, 192, 17, 17] 0
Conv2d-315 [-1, 192, 17, 17] 258,048
BatchNorm2d-316 [-1, 192, 17, 17] 384
ReLU-317 [-1, 192, 17, 17] 0
BasicConv2d-318 [-1, 192, 17, 17] 0
Conv2d-319 [-1, 192, 8, 8] 331,776
BatchNorm2d-320 [-1, 192, 8, 8] 384
ReLU-321 [-1, 192, 8, 8] 0
BasicConv2d-322 [-1, 192, 8, 8] 0
MaxPool2d-323 [-1, 768, 8, 8] 0
ReductionB-324 [-1, 1280, 8, 8] 0
Conv2d-325 [-1, 320, 8, 8] 409,600
BatchNorm2d-326 [-1, 320, 8, 8] 640
ReLU-327 [-1, 320, 8, 8] 0
BasicConv2d-328 [-1, 320, 8, 8] 0
Conv2d-329 [-1, 384, 8, 8] 491,520
BatchNorm2d-330 [-1, 384, 8, 8] 768
ReLU-331 [-1, 384, 8, 8] 0
BasicConv2d-332 [-1, 384, 8, 8] 0
Conv2d-333 [-1, 384, 8, 8] 442,368
BatchNorm2d-334 [-1, 384, 8, 8] 768
ReLU-335 [-1, 384, 8, 8] 0
BasicConv2d-336 [-1, 384, 8, 8] 0
Conv2d-337 [-1, 384, 8, 8] 442,368
BatchNorm2d-338 [-1, 384, 8, 8] 768
ReLU-339 [-1, 384, 8, 8] 0
BasicConv2d-340 [-1, 384, 8, 8] 0
Conv2d-341 [-1, 448, 8, 8] 573,440
BatchNorm2d-342 [-1, 448, 8, 8] 896
ReLU-343 [-1, 448, 8, 8] 0
BasicConv2d-344 [-1, 448, 8, 8] 0
Conv2d-345 [-1, 384, 8, 8] 1,548,288
BatchNorm2d-346 [-1, 384, 8, 8] 768
ReLU-347 [-1, 384, 8, 8] 0
BasicConv2d-348 [-1, 384, 8, 8] 0
Conv2d-349 [-1, 384, 8, 8] 442,368
BatchNorm2d-350 [-1, 384, 8, 8] 768
ReLU-351 [-1, 384, 8, 8] 0
BasicConv2d-352 [-1, 384, 8, 8] 0
Conv2d-353 [-1, 384, 8, 8] 442,368
BatchNorm2d-354 [-1, 384, 8, 8] 768
ReLU-355 [-1, 384, 8, 8] 0
BasicConv2d-356 [-1, 384, 8, 8] 0
AvgPool2d-357 [-1, 1280, 8, 8] 0
Conv2d-358 [-1, 192, 8, 8] 245,760
BatchNorm2d-359 [-1, 192, 8, 8] 384
ReLU-360 [-1, 192, 8, 8] 0
BasicConv2d-361 [-1, 192, 8, 8] 0
InceptionC-362 [-1, 2048, 8, 8] 0
Conv2d-363 [-1, 320, 8, 8] 655,360
BatchNorm2d-364 [-1, 320, 8, 8] 640
ReLU-365 [-1, 320, 8, 8] 0
BasicConv2d-366 [-1, 320, 8, 8] 0
Conv2d-367 [-1, 384, 8, 8] 786,432
BatchNorm2d-368 [-1, 384, 8, 8] 768
ReLU-369 [-1, 384, 8, 8] 0
BasicConv2d-370 [-1, 384, 8, 8] 0
Conv2d-371 [-1, 384, 8, 8] 442,368
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, 448, 8, 8] 917,504
BatchNorm2d-380 [-1, 448, 8, 8] 896
ReLU-381 [-1, 448, 8, 8] 0
BasicConv2d-382 [-1, 448, 8, 8] 0
Conv2d-383 [-1, 384, 8, 8] 1,548,288
BatchNorm2d-384 [-1, 384, 8, 8] 768
ReLU-385 [-1, 384, 8, 8] 0
BasicConv2d-386 [-1, 384, 8, 8] 0
Conv2d-387 [-1, 384, 8, 8] 442,368
BatchNorm2d-388 [-1, 384, 8, 8] 768
ReLU-389 [-1, 384, 8, 8] 0
BasicConv2d-390 [-1, 384, 8, 8] 0
Conv2d-391 [-1, 384, 8, 8] 442,368
BatchNorm2d-392 [-1, 384, 8, 8] 768
ReLU-393 [-1, 384, 8, 8] 0
BasicConv2d-394 [-1, 384, 8, 8] 0
AvgPool2d-395 [-1, 2048, 8, 8] 0
Conv2d-396 [-1, 192, 8, 8] 393,216
BatchNorm2d-397 [-1, 192, 8, 8] 384
ReLU-398 [-1, 192, 8, 8] 0
BasicConv2d-399 [-1, 192, 8, 8] 0
InceptionC-400 [-1, 2048, 8, 8] 0
Linear-401 [-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): 313.22
Params size (MB): 83.14
Estimated Total Size (MB): 397.38
----------------------------------------------------------------
InceptionV3(
(stem): Sequential(
(0): BasicConv2d(
(conv): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), bias=False)
(norm): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), bias=False)
(norm): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): BasicConv2d(
(conv): Conv2d(64, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(80, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): BasicConv2d(
(conv): Conv2d(80, 192, kernel_size=(3, 3), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(6): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(block1): Sequential(
(0): InceptionA(
(conv1): BasicConv2d(
(conv): Conv2d(192, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(192, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(48, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(48, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(192, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(192, 32, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(32, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(1): InceptionA(
(conv1): BasicConv2d(
(conv): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(256, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(48, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(48, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(2): InceptionA(
(conv1): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(288, 48, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(48, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(48, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
)
(block2): Sequential(
(0): ReductionA(
(conv1): BasicConv2d(
(conv): Conv2d(288, 384, kernel_size=(3, 3), stride=(2, 2), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(288, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(64, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(64, 96, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), bias=False)
(norm): BatchNorm2d(96, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(1): InceptionB(
(conv1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(128, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): BasicConv2d(
(conv): Conv2d(128, 128, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(128, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): BasicConv2d(
(conv): Conv2d(128, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(2): InceptionB(
(conv1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(3): InceptionB(
(conv1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): BasicConv2d(
(conv): Conv2d(160, 160, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(160, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): BasicConv2d(
(conv): Conv2d(160, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(4): InceptionB(
(conv1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
)
(block3): Sequential(
(0): ReductionB(
(conv1): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(192, 320, kernel_size=(3, 3), stride=(2, 2), bias=False)
(norm): BatchNorm2d(320, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv2): Sequential(
(0): BasicConv2d(
(conv): Conv2d(768, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(1, 7), stride=(1, 1), padding=(0, 3), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(7, 1), stride=(1, 1), padding=(3, 0), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): BasicConv2d(
(conv): Conv2d(192, 192, kernel_size=(3, 3), stride=(2, 2), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(1): InceptionC(
(conv1): BasicConv2d(
(conv): Conv2d(1280, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(320, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): BasicConv2d(
(conv): Conv2d(1280, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(1280, 448, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(448, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(448, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv3b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(1280, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(192, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
)
(2): InceptionC(
(conv1): BasicConv2d(
(conv): Conv2d(2048, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(320, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2): BasicConv2d(
(conv): Conv2d(2048, 384, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv2b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv3): Sequential(
(0): BasicConv2d(
(conv): Conv2d(2048, 448, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): BatchNorm2d(448, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(1): BasicConv2d(
(conv): Conv2d(448, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(conv3a): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(1, 3), stride=(1, 1), padding=(0, 1), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv3b): BasicConv2d(
(conv): Conv2d(384, 384, kernel_size=(3, 1), stride=(1, 1), padding=(1, 0), bias=False)
(norm): BatchNorm2d(384, eps=0.001, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(conv4): Sequential(
(0): AvgPool2d(kernel_size=3, stride=1, padding=1)
(1): BasicConv2d(
(conv): Conv2d(2048, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)
(norm): 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)
)
扫一扫
778
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
