resnet101网络结构图

以下是ResNet101的网络结构图：

Input: 224x224x3

Layer 1:
Convolutional layer (7x7 kernel, stride 2) -> Batch Normalization -> ReLU -> Max Pooling (3x3 kernel, stride 2)
Output: 56x56x64

Layer 2:
Convolutional layer (1x1 kernel, stride 1) -> Batch Normalization -> ReLU
Convolutional layer (3x3 kernel, stride 1) -> Batch Normalization -> ReLU
Convolutional layer (1x1 kernel, stride 1) -> Batch Normalization -> ReLU
Shortcut connection: Input -> Batch Normalization
Output: 56x56x256

Layer 3:
Convolutional layer (1x1 kernel, stride 1) -> Batch Normalization -> ReLU
Convolutional layer (3x3 kernel, stride 2) -> Batch Normalization -> ReLU
Convolutional layer (1x1 kernel, stride 1) -> Batch Normalization -> ReLU
Shortcut connection: Input -> Convolutional layer (1x1 kernel, stride 2) -> Batch Normalization
Output: 28x28x512

Layers 4-23:
Similar structure as Layer 3, but with increasing number of filters and decreasing spatial dimensions
Shortcut connection: Input -> Convolutional layer (1x1 kernel, appropriate stride) -> Batch Normalization
Output: 7x7x2048

Layer 24:
Average Pooling (7x7 kernel, stride 1)
Output: 1x1x2048

Layer 25:
Fully Connected layer (output size: number of classes)
Output: Number of classes