深度学习训练camp-第J9周：Inception v3算法实战与解析

最新推荐文章于 2025-05-02 19:08:58 发布
原创 最新推荐文章于 2025-05-02 19:08:58 发布 · 944 阅读 · 19 ·
本内容遵循CC 4.0 BY-SA版权协议
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
GEO检测
标签
#深度学习 #算法 #人工智能 #pytorch

📌 本周任务：

  1. 了解并学习InceptionV3相对于InceptionV1改进了哪些地方（重点）
  2. 使用Inception v3完成天气识别案例

Inception V3与V1之间的区别：
相对于InceptionV1（即GoogLeNet），InceptionV3在网络结构和训练策略上均进行了多方面的改进，旨在提升模型的表达能力、计算效率以及泛化性能。下面就主要改进点做详细说明：

  • 卷积核因式分解（Factorization into Smaller Convolutions）

  • 大卷积核分解：InceptionV1中某些模块采用了5×5卷积，但5×5卷积不仅参数量较大，而且计算量较高。InceptionV3将5×5卷积因式分解为连续的两个3×3卷积，既保持了相同的感受野，又显著降低了参数量和计算复杂度。

  • 异构卷积分解：在进一步优化方面，InceptionV3还将3×3卷积进一步分解为1×3和3×1的两个卷积操作。这种异构分解方法能够更精细地捕捉空间信息，同时降低计算代价，提高了模型效率。

  • 批归一化（Batch Normalization）的广泛应用

  • InceptionV3在更多卷积层后引入了批归一化技术。这一改进有助于稳定各层输出的分布，加速训练收敛，并在一定程度上起到正则化作用，从而提升了模型的泛化能力。相较于InceptionV1中对批归一化的应用较为有限，V3版本在此方面更为充分，整体训练过程更为稳健。

  • 标签平滑（Label Smoothing）正则化

  • 为缓解模型过拟合问题，InceptionV3在损失函数中引入了标签平滑技术。通过对目标标签进行平滑处理，减弱模型对训练数据中噪声的过度拟合，使得输出分布更为平滑，从而提高了模型在测试集上的泛化性能。

  • 网络结构的深度与宽度调整

  • 在保持计算资源合理的前提下，InceptionV3对网络的深度和宽度进行了优化调整。通过增加层数和调整Inception模块内部的结构，模型能够学习到更丰富、更细粒度的特征表示。这种结构上的优化有助于提高模型在复杂视觉任务中的表现。

  • 辅助分类器的改进

  • InceptionV1中通过辅助分类器（Auxiliary Classifiers）来缓解梯度消失问题，并起到一定的正则化效果。InceptionV3在保留这一设计思想的同时，对辅助分类器的结构和权重分配进行了精细化调整，使其与主分类器的协同训练效果更佳，从而整体上改善了模型性能。

  • 优化器与训练技巧的改进

  • 除了网络结构的改进，InceptionV3在训练过程中也采用了更为先进的优化方法（如RMSProp）以及针对性的数据预处理和数据增强技术。这些训练技巧的改进不仅加快了模型的收敛速度，还提升了模型在不同任务和数据集上的鲁棒性。

总结：InceptionV3相对于InceptionV1的主要改进体现在以下几个方面：

  • 通过将大卷积核（如5×5）因式分解为更小的卷积核组合（例如两个3×3或1×3与3×1的组合），有效降低了模型参数量和计算复杂度，同时提升了非线性特征的表达能力；
    在更大范围内应用批归一化技术，加速训练并提高稳定性； 引入标签平滑正则化，缓解过拟合问题；
    对网络结构进行深度与宽度的优化调整，以及辅助分类器的精细设计，进一步增强了模型的表达能力和训练效果；
    综合采用先进的优化器和训练策略，提高了整体模型性能。
    这些改进措施共同作用，使得InceptionV3在准确率、效率和鲁棒性等方面均优于其前身InceptionV1，成为计算机视觉任务中更为高效和准确的深度卷积神经网络架构。

1、搭建网络

1、Inception-A

在这里插入图片描述

class InceptionA(nn.Module):

    def __init__(self, in_channels, pool_features):
        super(InceptionA, self).__init__()
        self.branch1x1 = BasicConv2d(in_channels, 64, kernel_size=1) # 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)

2、Inception-B

在这里插入图片描述

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)

3、Inception-C

在这里插入图片描述

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)

4、Reduction-A

在这里插入图片描述

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)

5、Reduction-B

在这里插入图片描述

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)

6、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):
        # 17 x 17 x 768
        x = F.avg_pool2d(x, kernel_size=5, stride=3)
        # 5 x 5 x 768
        x = self.conv0(x)
        # 5 x 5 x 128
        x = self.conv1(x)
        # 1 x 1 x 768
        x = x.view(x.size(0), -1)
        # 768
        x = self.fc(x)
        # 1000
        return x

7、模型搭建

在这里插入图片描述

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: # 1
            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
        # 299 x 299 x 3
        x = self.Conv2d_1a_3x3(x)
        # 149 x 149 x 32
        x = self.Conv2d_2a_3x3(x)
        # 147 x 147 x 32
        x = self.Conv2d_2b_3x3(x)
        # 147 x 147 x 64
        x = F.max_pool2d(x, kernel_size=3, stride=2)
        # 73 x 73 x 64
        x = self.Conv2d_3b_1x1(x)
        # 73 x 73 x 80
        x = self.Conv2d_4a_3x3(x)
        # 71 x 71 x 192
        x = F.max_pool2d(x, kernel_size=3, stride=2)
        # 35 x 35 x 192
        x = self.Mixed_5b(x)
        # 35 x 35 x 256
        x = self.Mixed_5c(x)
        # 35 x 35 x 288
        x = self.Mixed_5d(x)
        # 35 x 35 x 288
        x = self.Mixed_6a(x)
        # 17 x 17 x 768
        x = self.Mixed_6b(x)
        # 17 x 17 x 768
        x = self.Mixed_6c(x)
        # 17 x 17 x 768
        x = self.Mixed_6d(x)
        # 17 x 17 x 768
        x = self.Mixed_6e(x)
        # 17 x 17 x 768
        if self.training and self.aux_logits:
            aux = self.AuxLogits(x)
        # 17 x 17 x 768
        x = self.Mixed_7a(x)
        # 8 x 8 x 1280
        x = self.Mixed_7b(x)
        # 8 x 8 x 2048
        x = self.Mixed_7c(x)
        # 8 x 8 x 2048
        x = F.avg_pool2d(x, kernel_size=8)
        # 1 x 1 x 2048
        x = F.dropout(x, training=self.training)
        # 1 x 1 x 2048
        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 x

2、网络总结

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
    
model = InceptionV3().to(device)
model

代码输出：

Using cuda device

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)
  )
  (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)
  )
  (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)
  )
  (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)
  )
  (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)
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
    (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)
    )
  )
  (fc): Linear(in_features=2048, out_features=1000, bias=True)
)

import torchsummary as summary
summary.summary(model, (3, 299, 299))

代码输出：

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1         [-1, 32, 149, 149]             864
       BatchNorm2d-2         [-1, 32, 149, 149]              64
       BasicConv2d-3         [-1, 32, 149, 149]               0
            Conv2d-4         [-1, 32, 147, 147]           9,216
       BatchNorm2d-5         [-1, 32, 147, 147]              64
       BasicConv2d-6         [-1, 32, 147, 147]               0
            Conv2d-7         [-1, 64, 147, 147]          18,432
       BatchNorm2d-8         [-1, 64, 147, 147]             128
       BasicConv2d-9         [-1, 64, 147, 147]               0
           Conv2d-10           [-1, 80, 73, 73]           5,120
      BatchNorm2d-11           [-1, 80, 73, 73]             160
      BasicConv2d-12           [-1, 80, 73, 73]               0
           Conv2d-13          [-1, 192, 71, 71]         138,240
      BatchNorm2d-14          [-1, 192, 71, 71]             384
      BasicConv2d-15          [-1, 192, 71, 71]               0
           Conv2d-16           [-1, 64, 35, 35]          12,288
      BatchNorm2d-17           [-1, 64, 35, 35]             128
      BasicConv2d-18           [-1, 64, 35, 35]               0
           Conv2d-19           [-1, 48, 35, 35]           9,216
      BatchNorm2d-20           [-1, 48, 35, 35]              96
      BasicConv2d-21           [-1, 48, 35, 35]               0
           Conv2d-22           [-1, 64, 35, 35]          76,800
      BatchNorm2d-23           [-1, 64, 35, 35]             128
      BasicConv2d-24           [-1, 64, 35, 35]               0
           Conv2d-25           [-1, 64, 35, 35]          12,288
      BatchNorm2d-26           [-1, 64, 35, 35]             128
      BasicConv2d-27           [-1, 64, 35, 35]               0
           Conv2d-28           [-1, 96, 35, 35]          55,296
      BatchNorm2d-29           [-1, 96, 35, 35]             192
      BasicConv2d-30           [-1, 96, 35, 35]               0
           Conv2d-31           [-1, 96, 35, 35]          82,944
      BatchNorm2d-32           [-1, 96, 35, 35]             192
      BasicConv2d-33           [-1, 96, 35, 35]               0
           Conv2d-34           [-1, 32, 35, 35]           6,144
      BatchNorm2d-35           [-1, 32, 35, 35]              64
      BasicConv2d-36           [-1, 32, 35, 35]               0
       InceptionA-37          [-1, 256, 35, 35]               0
           Conv2d-38           [-1, 64, 35, 35]          16,384
      BatchNorm2d-39           [-1, 64, 35, 35]             128
      BasicConv2d-40           [-1, 64, 35, 35]               0
           Conv2d-41           [-1, 48, 35, 35]          12,288
      BatchNorm2d-42           [-1, 48, 35, 35]              96
      BasicConv2d-43           [-1, 48, 35, 35]               0
           Conv2d-44           [-1, 64, 35, 35]          76,800
      BatchNorm2d-45           [-1, 64, 35, 35]             128
      BasicConv2d-46           [-1, 64, 35, 35]               0
           Conv2d-47           [-1, 64, 35, 35]          16,384
      BatchNorm2d-48           [-1, 64, 35, 35]             128
      BasicConv2d-49           [-1, 64, 35, 35]               0
           Conv2d-50           [-1, 96, 35, 35]          55,296
      BatchNorm2d-51           [-1, 96, 35, 35]             192
      BasicConv2d-52           [-1, 96, 35, 35]               0
           Conv2d-53           [-1, 96, 35, 35]          82,944
      BatchNorm2d-54           [-1, 96, 35, 35]             192
      BasicConv2d-55           [-1, 96, 35, 35]               0
           Conv2d-56           [-1, 64, 35, 35]          16,384
      BatchNorm2d-57           [-1, 64, 35, 35]             128
      BasicConv2d-58           [-1, 64, 35, 35]               0
       InceptionA-59          [-1, 288, 35, 35]               0
           Conv2d-60           [-1, 64, 35, 35]          18,432
      BatchNorm2d-61           [-1, 64, 35, 35]             128
      BasicConv2d-62           [-1, 64, 35, 35]               0
           Conv2d-63           [-1, 48, 35, 35]          13,824
      BatchNorm2d-64           [-1, 48, 35, 35]              96
      BasicConv2d-65           [-1, 48, 35, 35]               0
           Conv2d-66           [-1, 64, 35, 35]          76,800
      BatchNorm2d-67           [-1, 64, 35, 35]             128
      BasicConv2d-68           [-1, 64, 35, 35]               0
           Conv2d-69           [-1, 64, 35, 35]          18,432
      BatchNorm2d-70           [-1, 64, 35, 35]             128
      BasicConv2d-71           [-1, 64, 35, 35]               0
           Conv2d-72           [-1, 96, 35, 35]          55,296
      BatchNorm2d-73           [-1, 96, 35, 35]             192
      BasicConv2d-74           [-1, 96, 35, 35]               0
           Conv2d-75           [-1, 96, 35, 35]          82,944
      BatchNorm2d-76           [-1, 96, 35, 35]             192
      BasicConv2d-77           [-1, 96, 35, 35]               0
           Conv2d-78           [-1, 64, 35, 35]          18,432
      BatchNorm2d-79           [-1, 64, 35, 35]             128
      BasicConv2d-80           [-1, 64, 35, 35]               0
       InceptionA-81          [-1, 288, 35, 35]               0
           Conv2d-82          [-1, 384, 17, 17]         995,328
      BatchNorm2d-83          [-1, 384, 17, 17]             768
      BasicConv2d-84          [-1, 384, 17, 17]               0
           Conv2d-85           [-1, 64, 35, 35]          18,432
      BatchNorm2d-86           [-1, 64, 35, 35]             128
      BasicConv2d-87           [-1, 64, 35, 35]               0
           Conv2d-88           [-1, 96, 35, 35]          55,296
      BatchNorm2d-89           [-1, 96, 35, 35]             192
      BasicConv2d-90           [-1, 96, 35, 35]               0
           Conv2d-91           [-1, 96, 17, 17]          82,944
      BatchNorm2d-92           [-1, 96, 17, 17]             192
      BasicConv2d-93           [-1, 96, 17, 17]               0
       ReductionA-94          [-1, 768, 17, 17]               0
           Conv2d-95          [-1, 192, 17, 17]         147,456
      BatchNorm2d-96          [-1, 192, 17, 17]             384
      BasicConv2d-97          [-1, 192, 17, 17]               0
           Conv2d-98          [-1, 128, 17, 17]          98,304
      BatchNorm2d-99          [-1, 128, 17, 17]             256
     BasicConv2d-100          [-1, 128, 17, 17]               0
          Conv2d-101          [-1, 128, 17, 17]         114,688
     BatchNorm2d-102          [-1, 128, 17, 17]             256
     BasicConv2d-103          [-1, 128, 17, 17]               0
          Conv2d-104          [-1, 192, 17, 17]         172,032
     BatchNorm2d-105          [-1, 192, 17, 17]             384
     BasicConv2d-106          [-1, 192, 17, 17]               0
          Conv2d-107          [-1, 128, 17, 17]          98,304
     BatchNorm2d-108          [-1, 128, 17, 17]             256
     BasicConv2d-109          [-1, 128, 17, 17]               0
          Conv2d-110          [-1, 128, 17, 17]         114,688
     BatchNorm2d-111          [-1, 128, 17, 17]             256
     BasicConv2d-112          [-1, 128, 17, 17]               0
          Conv2d-113          [-1, 128, 17, 17]         114,688
     BatchNorm2d-114          [-1, 128, 17, 17]             256
     BasicConv2d-115          [-1, 128, 17, 17]               0
          Conv2d-116          [-1, 128, 17, 17]         114,688
     BatchNorm2d-117          [-1, 128, 17, 17]             256
     BasicConv2d-118          [-1, 128, 17, 17]               0
          Conv2d-119          [-1, 192, 17, 17]         172,032
     BatchNorm2d-120          [-1, 192, 17, 17]             384
     BasicConv2d-121          [-1, 192, 17, 17]               0
          Conv2d-122          [-1, 192, 17, 17]         147,456
     BatchNorm2d-123          [-1, 192, 17, 17]             384
     BasicConv2d-124          [-1, 192, 17, 17]               0
      InceptionB-125          [-1, 768, 17, 17]               0
          Conv2d-126          [-1, 192, 17, 17]         147,456
     BatchNorm2d-127          [-1, 192, 17, 17]             384
     BasicConv2d-128          [-1, 192, 17, 17]               0
          Conv2d-129          [-1, 160, 17, 17]         122,880
     BatchNorm2d-130          [-1, 160, 17, 17]             320
     BasicConv2d-131          [-1, 160, 17, 17]               0
          Conv2d-132          [-1, 160, 17, 17]         179,200
     BatchNorm2d-133          [-1, 160, 17, 17]             320
     BasicConv2d-134          [-1, 160, 17, 17]               0
          Conv2d-135          [-1, 192, 17, 17]         215,040
     BatchNorm2d-136          [-1, 192, 17, 17]             384
     BasicConv2d-137          [-1, 192, 17, 17]               0
          Conv2d-138          [-1, 160, 17, 17]         122,880
     BatchNorm2d-139          [-1, 160, 17, 17]             320
     BasicConv2d-140          [-1, 160, 17, 17]               0
          Conv2d-141          [-1, 160, 17, 17]         179,200
     BatchNorm2d-142          [-1, 160, 17, 17]             320
     BasicConv2d-143          [-1, 160, 17, 17]               0
          Conv2d-144          [-1, 160, 17, 17]         179,200
     BatchNorm2d-145          [-1, 160, 17, 17]             320
     BasicConv2d-146          [-1, 160, 17, 17]               0
          Conv2d-147          [-1, 160, 17, 17]         179,200
     BatchNorm2d-148          [-1, 160, 17, 17]             320
     BasicConv2d-149          [-1, 160, 17, 17]               0
          Conv2d-150          [-1, 192, 17, 17]         215,040
     BatchNorm2d-151          [-1, 192, 17, 17]             384
     BasicConv2d-152          [-1, 192, 17, 17]               0
          Conv2d-153          [-1, 192, 17, 17]         147,456
     BatchNorm2d-154          [-1, 192, 17, 17]             384
     BasicConv2d-155          [-1, 192, 17, 17]               0
      InceptionB-156          [-1, 768, 17, 17]               0
          Conv2d-157          [-1, 192, 17, 17]         147,456
     BatchNorm2d-158          [-1, 192, 17, 17]             384
     BasicConv2d-159          [-1, 192, 17, 17]               0
          Conv2d-160          [-1, 160, 17, 17]         122,880
     BatchNorm2d-161          [-1, 160, 17, 17]             320
     BasicConv2d-162          [-1, 160, 17, 17]               0
          Conv2d-163          [-1, 160, 17, 17]         179,200
     BatchNorm2d-164          [-1, 160, 17, 17]             320
     BasicConv2d-165          [-1, 160, 17, 17]               0
          Conv2d-166          [-1, 192, 17, 17]         215,040
     BatchNorm2d-167          [-1, 192, 17, 17]             384
     BasicConv2d-168          [-1, 192, 17, 17]               0
          Conv2d-169          [-1, 160, 17, 17]         122,880
     BatchNorm2d-170          [-1, 160, 17, 17]             320
     BasicConv2d-171          [-1, 160, 17, 17]               0
          Conv2d-172          [-1, 160, 17, 17]         179,200
     BatchNorm2d-173          [-1, 160, 17, 17]             320
     BasicConv2d-174          [-1, 160, 17, 17]               0
          Conv2d-175          [-1, 160, 17, 17]         179,200
     BatchNorm2d-176          [-1, 160, 17, 17]             320
     BasicConv2d-177          [-1, 160, 17, 17]               0
          Conv2d-178          [-1, 160, 17, 17]         179,200
     BatchNorm2d-179          [-1, 160, 17, 17]             320
     BasicConv2d-180          [-1, 160, 17, 17]               0
          Conv2d-181          [-1, 192, 17, 17]         215,040
     BatchNorm2d-182          [-1, 192, 17, 17]             384
     BasicConv2d-183          [-1, 192, 17, 17]               0
          Conv2d-184          [-1, 192, 17, 17]         147,456
     BatchNorm2d-185          [-1, 192, 17, 17]             384
     BasicConv2d-186          [-1, 192, 17, 17]               0
      InceptionB-187          [-1, 768, 17, 17]               0
          Conv2d-188          [-1, 192, 17, 17]         147,456
     BatchNorm2d-189          [-1, 192, 17, 17]             384
     BasicConv2d-190          [-1, 192, 17, 17]               0
          Conv2d-191          [-1, 192, 17, 17]         147,456
     BatchNorm2d-192          [-1, 192, 17, 17]             384
     BasicConv2d-193          [-1, 192, 17, 17]               0
          Conv2d-194          [-1, 192, 17, 17]         258,048
     BatchNorm2d-195          [-1, 192, 17, 17]             384
     BasicConv2d-196          [-1, 192, 17, 17]               0
          Conv2d-197          [-1, 192, 17, 17]         258,048
     BatchNorm2d-198          [-1, 192, 17, 17]             384
     BasicConv2d-199          [-1, 192, 17, 17]               0
          Conv2d-200          [-1, 192, 17, 17]         147,456
     BatchNorm2d-201          [-1, 192, 17, 17]             384
     BasicConv2d-202          [-1, 192, 17, 17]               0
          Conv2d-203          [-1, 192, 17, 17]         258,048
     BatchNorm2d-204          [-1, 192, 17, 17]             384
     BasicConv2d-205          [-1, 192, 17, 17]               0
          Conv2d-206          [-1, 192, 17, 17]         258,048
     BatchNorm2d-207          [-1, 192, 17, 17]             384
     BasicConv2d-208          [-1, 192, 17, 17]               0
          Conv2d-209          [-1, 192, 17, 17]         258,048
     BatchNorm2d-210          [-1, 192, 17, 17]             384
     BasicConv2d-211          [-1, 192, 17, 17]               0
          Conv2d-212          [-1, 192, 17, 17]         258,048
     BatchNorm2d-213          [-1, 192, 17, 17]             384
     BasicConv2d-214          [-1, 192, 17, 17]               0
          Conv2d-215          [-1, 192, 17, 17]         147,456
     BatchNorm2d-216          [-1, 192, 17, 17]             384
     BasicConv2d-217          [-1, 192, 17, 17]               0
      InceptionB-218          [-1, 768, 17, 17]               0
          Conv2d-219          [-1, 192, 17, 17]         147,456
     BatchNorm2d-220          [-1, 192, 17, 17]             384
     BasicConv2d-221          [-1, 192, 17, 17]               0
          Conv2d-222            [-1, 320, 8, 8]         552,960
     BatchNorm2d-223            [-1, 320, 8, 8]             640
     BasicConv2d-224            [-1, 320, 8, 8]               0
          Conv2d-225          [-1, 192, 17, 17]         147,456
     BatchNorm2d-226          [-1, 192, 17, 17]             384
     BasicConv2d-227          [-1, 192, 17, 17]               0
          Conv2d-228          [-1, 192, 17, 17]         258,048
     BatchNorm2d-229          [-1, 192, 17, 17]             384
     BasicConv2d-230          [-1, 192, 17, 17]               0
          Conv2d-231          [-1, 192, 17, 17]         258,048
     BatchNorm2d-232          [-1, 192, 17, 17]             384
     BasicConv2d-233          [-1, 192, 17, 17]               0
          Conv2d-234            [-1, 192, 8, 8]         331,776
     BatchNorm2d-235            [-1, 192, 8, 8]             384
     BasicConv2d-236            [-1, 192, 8, 8]               0
      ReductionB-237           [-1, 1280, 8, 8]               0
          Conv2d-238            [-1, 320, 8, 8]         409,600
     BatchNorm2d-239            [-1, 320, 8, 8]             640
     BasicConv2d-240            [-1, 320, 8, 8]               0
          Conv2d-241            [-1, 384, 8, 8]         491,520
     BatchNorm2d-242            [-1, 384, 8, 8]             768
     BasicConv2d-243            [-1, 384, 8, 8]               0
          Conv2d-244            [-1, 384, 8, 8]         442,368
     BatchNorm2d-245            [-1, 384, 8, 8]             768
     BasicConv2d-246            [-1, 384, 8, 8]               0
          Conv2d-247            [-1, 384, 8, 8]         442,368
     BatchNorm2d-248            [-1, 384, 8, 8]             768
     BasicConv2d-249            [-1, 384, 8, 8]               0
          Conv2d-250            [-1, 448, 8, 8]         573,440
     BatchNorm2d-251            [-1, 448, 8, 8]             896
     BasicConv2d-252            [-1, 448, 8, 8]               0
          Conv2d-253            [-1, 384, 8, 8]       1,548,288
     BatchNorm2d-254            [-1, 384, 8, 8]             768
     BasicConv2d-255            [-1, 384, 8, 8]               0
          Conv2d-256            [-1, 384, 8, 8]         442,368
     BatchNorm2d-257            [-1, 384, 8, 8]             768
     BasicConv2d-258            [-1, 384, 8, 8]               0
          Conv2d-259            [-1, 384, 8, 8]         442,368
     BatchNorm2d-260            [-1, 384, 8, 8]             768
     BasicConv2d-261            [-1, 384, 8, 8]               0
          Conv2d-262            [-1, 192, 8, 8]         245,760
     BatchNorm2d-263            [-1, 192, 8, 8]             384
     BasicConv2d-264            [-1, 192, 8, 8]               0
      InceptionC-265           [-1, 2048, 8, 8]               0
          Conv2d-266            [-1, 320, 8, 8]         655,360
     BatchNorm2d-267            [-1, 320, 8, 8]             640
     BasicConv2d-268            [-1, 320, 8, 8]               0
          Conv2d-269            [-1, 384, 8, 8]         786,432
     BatchNorm2d-270            [-1, 384, 8, 8]             768
     BasicConv2d-271            [-1, 384, 8, 8]               0
          Conv2d-272            [-1, 384, 8, 8]         442,368
     BatchNorm2d-273            [-1, 384, 8, 8]             768
     BasicConv2d-274            [-1, 384, 8, 8]               0
          Conv2d-275            [-1, 384, 8, 8]         442,368
     BatchNorm2d-276            [-1, 384, 8, 8]             768
     BasicConv2d-277            [-1, 384, 8, 8]               0
          Conv2d-278            [-1, 448, 8, 8]         917,504
     BatchNorm2d-279            [-1, 448, 8, 8]             896
     BasicConv2d-280            [-1, 448, 8, 8]               0
          Conv2d-281            [-1, 384, 8, 8]       1,548,288
     BatchNorm2d-282            [-1, 384, 8, 8]             768
     BasicConv2d-283            [-1, 384, 8, 8]               0
          Conv2d-284            [-1, 384, 8, 8]         442,368
     BatchNorm2d-285            [-1, 384, 8, 8]             768
     BasicConv2d-286            [-1, 384, 8, 8]               0
          Conv2d-287            [-1, 384, 8, 8]         442,368
     BatchNorm2d-288            [-1, 384, 8, 8]             768
     BasicConv2d-289            [-1, 384, 8, 8]               0
          Conv2d-290            [-1, 192, 8, 8]         393,216
     BatchNorm2d-291            [-1, 192, 8, 8]             384
     BasicConv2d-292            [-1, 192, 8, 8]               0
      InceptionC-293           [-1, 2048, 8, 8]               0
          Linear-294                 [-1, 1000]       2,049,000
================================================================
Total params: 23,834,568
Trainable params: 23,834,568
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.02
Forward/backward pass size (MB): 224.12
Params size (MB): 90.92
Estimated Total Size (MB): 316.07
----------------------------------------------------------------

3、用于天气预测

数据预处理部分每次都差不多了，我们这次省略。直接给出训练函数和测试函数以及训练过程。

import torch
import torch.nn as nn
import torch.nn.functional as F
import copy
from torch.optim.lr_scheduler import ReduceLROnPlateau

def train(dataloader, model, loss_fn, optimizer):
    model.train()  # 进入训练模式
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    
    total_loss = 0.0
    correct = 0
    
    for x, y in dataloader:
        x, y = x.to(device), y.to(device)
        
        optimizer.zero_grad()
        outputs = model(x)
        # 判断是否存在辅助分类器输出
        if isinstance(outputs, tuple):
            pred, aux = outputs
            main_loss = loss_fn(pred, y)
            aux_loss = loss_fn(aux, y)
            loss = main_loss + 0.6 * aux_loss
        else:
            pred = outputs
            loss = loss_fn(pred, y)
            
        loss.backward()
        optimizer.step()
        
        total_loss += loss.item()
        correct += (pred.argmax(dim=1) == y).type(torch.float).sum().item()
    
    avg_loss = total_loss / num_batches
    accuracy = correct / size
    return accuracy, avg_loss

def test(dataloader, model, loss_fn):
    model.eval()  # 进入评估模式
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    
    total_loss = 0.0
    correct = 0
    
    with torch.no_grad():
        for x, y in dataloader:
            x, y = x.to(device), y.to(device)
            outputs = model(x)
            # 若返回元组，则取主输出部分
            if isinstance(outputs, tuple):
                pred = outputs[0]
            else:
                pred = outputs
                
            loss = loss_fn(pred, y)
            total_loss += loss.item()
            correct += (pred.argmax(dim=1) == y).type(torch.float).sum().item()
    
    avg_loss = total_loss / num_batches
    accuracy = correct / size
    return accuracy, avg_loss


optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=0.1, patience=5, verbose=True)
loss_fn = nn.CrossEntropyLoss()

epochs = 32

# 用于记录每个 epoch 的损失和准确率
train_loss_list = []
train_acc_list = []
test_loss_list = []
test_acc_list = []

best_acc = 0.0

for epoch in range(epochs):
    # 训练阶段
    train_acc, train_loss = train(train_dl, model, loss_fn, optimizer)
    
    # 测试阶段
    test_acc, test_loss = test(test_dl, model, loss_fn)
    
    # 根据测试损失调整学习率
    scheduler.step(test_loss)
    
    # 保存最佳模型
    if test_acc > best_acc:
        best_acc = test_acc
        best_model = copy.deepcopy(model)
    
    train_acc_list.append(train_acc)
    train_loss_list.append(train_loss)
    test_acc_list.append(test_acc)
    test_loss_list.append(test_loss)
    
    # 获取当前学习率
    current_lr = optimizer.param_groups[0]['lr']
    print(f"Epoch:{epoch+1:2d}, Train_acc:{train_acc*100:.1f}%, Train_loss:{train_loss:.3f}, "
          f"Test_acc:{test_acc*100:.1f}%, Test_loss:{test_loss:.3f}, Lr:{current_lr:.2E}")

# 保存最佳模型
PATH = './best_model.pth'
torch.save(best_model.state_dict(), PATH)

print('Training Completed. Best Test Accuracy: {:.1f}%'.format(best_acc * 100))

代码输出：

Epoch: 1, Train_acc:67.0%, Train_loss:2.894, Test_acc:30.2%, Test_loss:6.561, Lr:1.00E-04
Epoch: 2, Train_acc:86.5%, Train_loss:0.577, Test_acc:90.6%, Test_loss:0.541, Lr:1.00E-04
Epoch: 3, Train_acc:89.1%, Train_loss:0.412, Test_acc:87.6%, Test_loss:0.610, Lr:1.00E-04
Epoch: 4, Train_acc:89.5%, Train_loss:0.393, Test_acc:86.1%, Test_loss:0.519, Lr:1.00E-04
Epoch: 5, Train_acc:92.0%, Train_loss:0.320, Test_acc:90.6%, Test_loss:0.481, Lr:1.00E-04
Epoch: 6, Train_acc:93.3%, Train_loss:0.208, Test_acc:92.1%, Test_loss:0.359, Lr:1.00E-04
Epoch: 7, Train_acc:94.2%, Train_loss:0.156, Test_acc:94.6%, Test_loss:0.230, Lr:1.00E-04
Epoch: 8, Train_acc:94.3%, Train_loss:0.167, Test_acc:90.6%, Test_loss:0.350, Lr:1.00E-04
Epoch: 9, Train_acc:95.6%, Train_loss:0.150, Test_acc:92.6%, Test_loss:0.287, Lr:1.00E-04
Epoch:10, Train_acc:93.6%, Train_loss:0.152, Test_acc:93.1%, Test_loss:0.248, Lr:1.00E-04
Epoch:11, Train_acc:95.0%, Train_loss:0.155, Test_acc:94.6%, Test_loss:0.187, Lr:1.00E-04
Epoch:12, Train_acc:94.5%, Train_loss:0.171, Test_acc:92.1%, Test_loss:0.307, Lr:1.00E-04
Epoch:13, Train_acc:97.3%, Train_loss:0.088, Test_acc:91.1%, Test_loss:0.410, Lr:1.00E-04
Epoch:14, Train_acc:94.8%, Train_loss:0.126, Test_acc:92.1%, Test_loss:0.277, Lr:1.00E-04
Epoch:15, Train_acc:95.4%, Train_loss:0.157, Test_acc:90.6%, Test_loss:0.283, Lr:1.00E-04
Epoch:16, Train_acc:95.3%, Train_loss:0.124, Test_acc:96.5%, Test_loss:0.234, Lr:1.00E-04
Epoch:17, Train_acc:99.0%, Train_loss:0.050, Test_acc:95.0%, Test_loss:0.166, Lr:1.00E-04
Epoch:18, Train_acc:98.6%, Train_loss:0.052, Test_acc:94.1%, Test_loss:0.213, Lr:1.00E-04
Epoch:19, Train_acc:98.5%, Train_loss:0.053, Test_acc:93.6%, Test_loss:0.210, Lr:1.00E-04
Epoch:20, Train_acc:97.3%, Train_loss:0.093, Test_acc:96.5%, Test_loss:0.183, Lr:1.00E-04
Epoch:21, Train_acc:98.3%, Train_loss:0.051, Test_acc:94.6%, Test_loss:0.184, Lr:1.00E-04
Epoch:22, Train_acc:98.3%, Train_loss:0.052, Test_acc:94.1%, Test_loss:0.199, Lr:1.00E-04
Epoch:23, Train_acc:98.5%, Train_loss:0.067, Test_acc:95.5%, Test_loss:0.325, Lr:1.00E-05
Epoch:24, Train_acc:97.6%, Train_loss:0.071, Test_acc:96.5%, Test_loss:0.265, Lr:1.00E-05
Epoch:25, Train_acc:99.5%, Train_loss:0.016, Test_acc:96.0%, Test_loss:0.287, Lr:1.00E-05
Epoch:26, Train_acc:99.5%, Train_loss:0.023, Test_acc:96.5%, Test_loss:0.201, Lr:1.00E-05
Epoch:27, Train_acc:99.2%, Train_loss:0.021, Test_acc:96.5%, Test_loss:0.254, Lr:1.00E-05
Epoch:28, Train_acc:100.0%, Train_loss:0.011, Test_acc:96.5%, Test_loss:0.178, Lr:1.00E-05
Epoch:29, Train_acc:99.4%, Train_loss:0.017, Test_acc:96.5%, Test_loss:0.171, Lr:1.00E-06
Epoch:30, Train_acc:99.4%, Train_loss:0.023, Test_acc:96.0%, Test_loss:0.173, Lr:1.00E-06
Epoch:31, Train_acc:99.5%, Train_loss:0.050, Test_acc:96.5%, Test_loss:0.190, Lr:1.00E-06
Epoch:32, Train_acc:99.7%, Train_loss:0.017, Test_acc:96.5%, Test_loss:0.178, Lr:1.00E-06
Training Completed. Best Test Accuracy: 96.5%

4、数据可视化

import matplotlib.pyplot as plt
epochs_range = range(epochs)

plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)

plt.plot(epochs_range, train_acc_list, label='Training Accuracy')
plt.plot(epochs_range, test_acc_list, 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_list, label='Training Loss')
plt.plot(epochs_range, test_loss_list, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()

代码输出：
在这里插入图片描述

5、验证集准确性：

def validate(dataloader, model):
    model.eval()
    size = len(dataloader.dataset)
    num_batches = len(dataloader)

    validate_acc = 0

    for x, y in dataloader:
        x, y = x.to(device), y.to(device)

        pred = model(x)

        validate_acc += (pred.argmax(1) == y).type(torch.float).sum().item()

    validate_acc /= size

    return validate_acc


# 计算验证集准确率
validate_acc = validate(validate_dl, best_model)
print(f"Validation Accuracy: {validate_acc:.2%}")

Validation Accuracy: 94.12%
iracole
关注
第39步 深度学习图像识别：Inception V3建模（Tensorflow）
Jet4505的博客
06-21 3100
第39步 深度学习图像识别：Inception V3建模（Tensorflow）
迁移学习---inceptionV3
笼中鸟
04-22 5124
一.迁移学习的概述 迁移学习是指将一个在某领域完成学习任务的网络应用到新的领域来进行学习任务。 基于卷积网络实现迁移学习 对于已经训练完毕的网络模型来说，通常该网络模型的前几层学习到的是通用特征，随着网络层次的加深，更深次的网络层更偏重于学习特定的特征，因此可将通用特征迁移到其他领域。对于深度卷积网络来说，经过预训练的网络模型可以实现网络结构与参数信息的分离，在保证网络结构一致的前提下，便可以利用经过预训练得到的权重参数初始化新的网络，这种方式可以极大地减少训练时间。 具体步骤如下： ①选定基于I
Tensorflow框架：InceptionV3网络概念及实现
瞳瞳瞳呀
06-10 3414
卷积神经网络迁移学习-Inception • 有论文依据表明可以保留训练好的inception模型中所有卷积层的参数，只替换最后一层全连接层。在最后 这一层全连接层之前的网络称为瓶颈层。 • 原理：在训练好的inception模型中，因为将瓶颈层的输出再通过一个单层的全连接层，神经网络可以很好 的区分1000种类别的图像，所以可以认为瓶颈层输出的节点向量可以被作为任何图像的一个更具有表达能 力的特征向量。于是在新的数据集上可以直接利用这个训练好的神经网络对图像进行特征提取，然后将提 取得到的特征向量作为输入
Inception V3
热门推荐
chairon的博客
08-06 2万+
目录0 回顾1 介绍2 设计原则3 大filter size卷积的分解3.1 分解为小卷积3.2 分解为非对称卷积4 辅助分类器的效用5 feature map的size的高效减小6 Inception v37. Label Smoothing模型正则8. 在低分辨率输入情况下的性能9. 实验中的结果和对比10.结论 0 回顾 Inception V1：主要提出了多分支（多分辨率的filter组合）的网络 Inception V2： 主要提出了BN层，提高网络性能（减少梯度消失和爆炸、防止过拟合
【图像分类】【深度学习】【Pytorch版本】GoogLeNet(InceptionV3)模型算法详解
yangyu0515的博客
11-15 6734
【图像分类】【深度学习】【Pytorch版本】GoogLeNet(InceptionV3)模型算法详解
深度学习训练camp：第J8周：Inception v1算法实战与解析
iracole的博客
01-20 813
经典神经网络论文超详细解读（三）——GoogLeNetInceptionV1学习笔记（翻译＋精读+代码复现）在背景部分已经十分详细了。GoogLeNetInceptionV1的最大特点是其Inception模块，该模块通过并行的卷积和池化操作来提取多尺度特征。具体来说，Inception模块允许网络在同一层中使用不同大小的卷积核（如1x1、3x3和5x5），以及最大池化，从而增强了模型的表达能力和灵活性。
深度学习第J9周：Inception v3算法实战与解析
m0_62237233的博客
05-14 6181
总结来说，Inception v3是一种深度卷积神经网络，其主要特点包括更深的网络结构、使用Factorized Convolutions、添加Batch Normalization、引入辅助分类器以及使用基于RMSProp的优化器进行训练。相对于Inception v1的Inception Module结构，Inception v3在卷积操作上做出了改动，使用两个3x3的卷积代替一个5x5的卷积以提高计算速度，并将n x n的卷积核尺寸分解为1 x n和n x 1两个卷积。
深度学习基础--CNN经典网络之InceptionV3详解与复现(pytorch)
羊小猪~~的博客
04-18 3261
是谷歌在2015年提出，是的进阶版，对于Inception系列网络来说，他是当时第一个在100层卷积网络却依然可以取得好效果的网络(ResNet还没有提出来)，相比于更深入得网络结构，在中，包含了48层卷积网络，这可以提取出更多特征，从而获得更好成果；使用分解卷积，将较大的卷积核分解为多个较小的卷积核，在保持良好性能的情况下，依然降低了网络参数量，减少计算复杂度；使用BN层，
Inception v3
kui9702的博客
03-07 6272
Inception v3 2014年ImageNet竞赛的冠军Inception-v1，又名GoogLeNet。Inception v1的特点：模块增加网络的宽度。将模型的的输入经过几种卷积的计算,以concat方式连接。 Inception v2，在v1版本上改进2个方向： 引入BN层 模型在计算过程中，会先对输入进行归一化 Inception v3，在之前的基础上增加： 将大卷积分解成小卷积，使得在感受野不变的情况下，减少参数的计算量 max pooling层在下采样会导致信息损失大，于是设
【精读AI论文】inceptionV3 （Rethinking the Inception Architecture for Computer Vision）
深度不学习！！的博客
04-21 7590
文章目录前言Abstract（摘要）Introduction（引言）General Design Principles（通用设计原则）原则一：避免过度的降维或者收缩特征 尤其在网络浅层。原则二：特征越多，收敛越快。原则三：3 * 3和5 * 5的卷积核卷积之前可以用1 * 1 的卷积核降维，信息不会丢失。原则四：均衡网络中的深度和宽度。Factorizing Convolutions with Large Filter Size（分解大卷积核 ）分解方法一：分解方法二： 前言 今天看一下inception-
InceptionV3代码实现（Pytorch）
Yophan的博客
07-18 2698
Inception网络是CNN发展史上一个重要的里程碑。在Inception出现之前，大部分流行CNN仅仅是把卷积层堆叠得越来越多，使网络越来越深，以此希望能够得到更好的性能。
cnn之inception-v3模型结构与参数浅析
qq_39691463的博客
08-14 1万+
背景总论 接上篇，在对LeNet有一定理解之后，我们开始谈谈CNN里面另一个经典的模型--inception-v3 笔者在勉强读完原文之后，不得不佩服原作者对cnn研究之深入，其对cnn整体架构，参数的 调优，训练的优化，以及整个模型的优化都提出了很多建设性的意见。值得我们学习。前事不 忘后事之师，我们开始吧！ 1. 论文背景 整个论文的思路主要是为了解决一个问题–如何处理网络更宽更窄带来的问题。 （1）：参数过多而训练数据不足的时候同意导致过拟合（如何减小参数） （2）：当网络越来越深，带来的
P20：Inception v3算法实战与解析
最新发布
e_motion的博客
05-02 1420
Inception v3是Google团队在2015年提出的第三代Inception模型，属于卷积神经网络（CNN）的经典架构之一，主要用于图像分类任务。它在Inception v1（即GoogLeNet）的基础上进行了多项优化，显著提升了模型性能和计算效率。
卷积神经网络的网络结构——Inception V3
小花生的博客
06-11 1万+
《Rethinking the Inception Architecture for Computer Vision》 2015，Google，Inception V3 1.基于大滤波器尺寸分解卷积 GoogLeNet性能优异很大程度在于使用了降维。降维可以看做卷积网络的因式分解。例如1x1卷积层后跟着3x3卷积层。在网络角度看，激活层的输出是高相关的；因此在聚合前进行降维，可以得到类似...
inceptionv3
tjl521314_21的博客
04-29 890
Inception v3是谷歌研究团队提出的深度卷积神经网络架构，通过引入多尺度的Inception模块和辅助分类器等技术，有效解决了深度网络训练中的梯度消失和模型参数过多的问题。该网络利用不同大小的卷积核并行处理图像信息，结合Batch Normalization和全局平均池化等优化策略，以较少的参数实现了优异的图像分类和识别性能，成为了深度学习图像处理领域的重要里程碑之一。下面分别进行介绍。
深度学习网络篇——Inception v3
PRIS-SCMonkey的博客
12-12 3369
文章目录先夸一夸我们的GoogLeNetInception v3的薅羊毛顺序第一部分 总体设计原则1、避免表达的瓶颈，特别是在网络前面的部分2、高维度特征更适合在网络局部中处理3、在较低维度的输入上进行空间聚合，不会降低网络表示能力4、平衡网络的宽度和深度第二部分 分解大卷积核1、大卷积分解为小卷积2、非对称卷积的空间分解第三部分 使用辅助分类器第四部分 有效网格尺寸缩减第五部分 Inceptio...
Inception v3算法实战与解析
weixin_44523812的博客
03-21 1047
学习了Inception v3模型的框架结构。
Inception v1、v2、v3
蓬莱道人的博客
07-15 1763
1、Incention V1（CVPR2015）：  （1）网络结构           GooLeNet的提出主要为为了解决这个问题，即通过增加网络层数量或者增加每层网络的宽度来提高准确率时有很多缺点，第一个是容易引起过拟合，第二个是造成计算资源极大增加。而解决这个问题的方法通常是让全连接层甚至卷积层变得稀疏化，这种方法不仅有仿生系统基础，而且这也有更坚实的理论基础的优势，即如果数据集的概...
20_Inception V3深度学习图像分类算法
https://github.com/foxpup11?tab=repositories
07-09 2182
https://www.bilibili.com/video/BV1FM4y157N2/?spm_id_from=333.999.0.0&vd_source=7dace3632125a1ef7fd32c285eb2fbac
J9 - Inception v3算法
Loser
03-15 1402
通过对InceptionV1和InceptionV3的对比可以发现，InceptionV1的模型结构十分简单，InceptionV3模型的结构就复杂了很多。然而InceptionV3的性能在InceptionV1的基础上略有提升，但不不是非常明显。是并行1xn和nx1的卷积来替换nxn卷积，在精度不变的情况下可以降低很多参数量和计算量。
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值