From 19974d761a142fd74651fe022cd157251791b6c3 Mon Sep 17 00:00:00 2001 From: guanjihuan Date: Wed, 3 Apr 2024 16:22:28 +0800 Subject: [PATCH] update --- .../data_size_after_convolution.py | 77 +++++++++++++++++++ .../data_size_after_pooling.py | 30 ++++++++ 2 files changed, 107 insertions(+) create mode 100644 2024.04.03_data_size_after_convolution_and_pooling/data_size_after_convolution.py create mode 100644 2024.04.03_data_size_after_convolution_and_pooling/data_size_after_pooling.py diff --git a/2024.04.03_data_size_after_convolution_and_pooling/data_size_after_convolution.py b/2024.04.03_data_size_after_convolution_and_pooling/data_size_after_convolution.py new file mode 100644 index 0000000..99923b2 --- /dev/null +++ b/2024.04.03_data_size_after_convolution_and_pooling/data_size_after_convolution.py @@ -0,0 +1,77 @@ +""" +This code is supported by the website: https://www.guanjihuan.com +The newest version of this code is on the web page: https://www.guanjihuan.com/archives/39320 +""" + +import torch +input_data = torch.randn(1, 1, 28, 28) + +print('【有填充的情况，不同卷积核大小对输出数据的维度的影响】') +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=1, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为1）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=2, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为2）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为3）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=4, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为4）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=5, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为5）：", output_data.shape) + +print() +print('【无填充的情况，不同卷积核大小对输出数据的维度的影响】') +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=1, stride=1, padding=0, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为1）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=2, stride=1, padding=0, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为2）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=0, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为3）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=4, stride=1, padding=0, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为4）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=5, stride=1, padding=0, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为5）：", output_data.shape) + +print() +print('【不同步幅对输出数据的维度的影响（以卷积核大小为3，有填充的情况为例）】') +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（步幅为1）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=2, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（步幅为2）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=3, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（步幅为3）：", output_data.shape) + +print() +print('【不同扩张率对输出数据的维度的影响（以步幅为1，有填充的情况为例）】') +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为3，扩张率为1）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=1, dilation=2) +output_data = conv_layer(input_data) +print("维度（卷积核大小为3，扩张率为2）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=1, dilation=3) +output_data = conv_layer(input_data) +print("维度（卷积核大小为3，扩张率为3）：", output_data.shape) + +print() +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=4, stride=1, padding=1, dilation=1) +output_data = conv_layer(input_data) +print("维度（卷积核大小为4，扩张率为1）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=4, stride=1, padding=1, dilation=2) +output_data = conv_layer(input_data) +print("维度（卷积核大小为4，扩张率为2）：", output_data.shape) +conv_layer = torch.nn.Conv2d(in_channels=1, out_channels=1, kernel_size=4, stride=1, padding=1, dilation=3) +output_data = conv_layer(input_data) +print("维度（卷积核大小为4，扩张率为3）：", output_data.shape) \ No newline at end of file diff --git a/2024.04.03_data_size_after_convolution_and_pooling/data_size_after_pooling.py b/2024.04.03_data_size_after_convolution_and_pooling/data_size_after_pooling.py new file mode 100644 index 0000000..7ac2ad2 --- /dev/null +++ b/2024.04.03_data_size_after_convolution_and_pooling/data_size_after_pooling.py @@ -0,0 +1,30 @@ +""" +This code is supported by the website: https://www.guanjihuan.com +The newest version of this code is on the web page: https://www.guanjihuan.com/archives/39320 +""" + +import torch +input_data = torch.randn(1, 1, 28, 28) + +print('【不同卷积核大小对输出数据的维度的影响】') +max_pool = torch.nn.MaxPool2d(kernel_size=1, stride=1) +output_data = max_pool(input_data) +print("维度（卷积核大小为1）：", output_data.shape) +max_pool = torch.nn.MaxPool2d(kernel_size=2, stride=1) +output_data = max_pool(input_data) +print("维度（卷积核大小为2）：", output_data.shape) +max_pool = torch.nn.MaxPool2d(kernel_size=3, stride=1) +output_data = max_pool(input_data) +print("维度（卷积核大小为3）：", output_data.shape) + +print() +print('【不同步幅对输出数据的维度的影响（以卷积核大小为2为例）】') +max_pool = torch.nn.MaxPool2d(kernel_size=2, stride=1) +output_data = max_pool(input_data) +print("维度（步幅为1）：", output_data.shape) +max_pool = torch.nn.MaxPool2d(kernel_size=2, stride=2) +output_data = max_pool(input_data) +print("维度（步幅为2）：", output_data.shape) +max_pool = torch.nn.MaxPool2d(kernel_size=2, stride=3) +output_data = max_pool(input_data) +print("维度（步幅为3）：", output_data.shape) \ No newline at end of file