查看torchinfo的网络结构输出
使用torchinfo，设置batchsize和图片大小就行。
我这里是导入了一个网络结构。可以自己去找。
from network.classifier import *
# model = MesoInception4()
# print(model)
from torchinfo import summary
model = MesoInception4()
batch_size = 128
summary(model, input_size=(batch_size, 3, 256, 256))
输出
==========================================================================================
Layer (type:depth-idx) Output Shape Param #
==========================================================================================
MesoInception4 [128, 2] --
├─Conv2d: 1-1 [128, 1, 256, 256] 3
├─Conv2d: 1-2 [128, 4, 256, 256] 12
├─Conv2d: 1-3 [128, 4, 256, 256] 144
├─Conv2d: 1-4 [128, 4, 256, 256] 12
├─Conv2d: 1-5 [128, 4, 256, 256] 144
├─Conv2d: 1-6 [128, 2, 256, 256] 6
├─Conv2d: 1-7 [128, 2, 256, 256] 36
├─BatchNorm2d: 1-8 [128, 11, 256, 256] 22
├─MaxPool2d: 1-9 [128, 11, 128, 128] --
├─Conv2d: 1-10 [128, 2, 128, 128] 22
├─Conv2d: 1-11 [128, 4, 128, 128] 44
├─Conv2d: 1-12 [128, 4, 128, 128] 144
├─Conv2d: 1-13 [128, 4, 128, 128] 44
├─Conv2d: 1-14 [128, 4, 128, 128] 144
├─Conv2d: 1-15 [128, 2, 128, 128] 22
├─Conv2d: 1-16 [128, 2, 128, 128] 36
├─BatchNorm2d: 1-17 [128, 12, 128, 128] 24
├─MaxPool2d: 1-18 [128, 12, 64, 64] --
├─Conv2d: 1-19 [128, 16, 64, 64] 4,800
├─ReLU: 1-20 [128, 16, 64, 64] --
├─BatchNorm2d: 1-21 [128, 16, 64, 64] 32
├─MaxPool2d: 1-22 [128, 16, 32, 32] --
├─Conv2d: 1-23 [128, 16, 32, 32] 6,400
├─ReLU: 1-24 [128, 16, 32, 32] --
├─BatchNorm2d: 1-25 [128, 16, 32, 32] (recursive)
├─MaxPool2d: 1-26 [128, 16, 8, 8] --
├─Dropout2d: 1-27 [128, 1024] --
├─Linear: 1-28 [128, 16] 16,400
├─LeakyReLU: 1-29 [128, 16] --
├─Dropout2d: 1-30 [128, 16] --
├─Linear: 1-31 [128, 2] 34
==========================================================================================
Total params: 28,525
Trainable params: 28,525
Non-trainable params: 0
Total mult-adds (G): 7.31
==========================================================================================
Input size (MB): 100.66
Forward/backward pass size (MB): 2868.92
Params size (MB): 0.11
Estimated Total Size (MB): 2969.70
==========================================================================================
直接使用print是这样，可以结合使用看。能画出图最好了。
MesoInception4(
(Incption1_conv1): Conv2d(3, 1, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption1_conv2_1): Conv2d(3, 4, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption1_conv2_2): Conv2d(4, 4, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(Incption1_conv3_1): Conv2d(3, 4, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption1_conv3_2): Conv2d(4, 4, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)
(Incption1_conv4_1): Conv2d(3, 2, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption1_conv4_2): Conv2d(2, 2, kernel_size=(3, 3), stride=(1, 1), padding=(3, 3), dilation=(3, 3), bias=False)
(Incption1_bn): BatchNorm2d(11, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(Incption2_conv1): Conv2d(11, 2, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption2_conv2_1): Conv2d(11, 4, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption2_conv2_2): Conv2d(4, 4, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(Incption2_conv3_1): Conv2d(11, 4, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption2_conv3_2): Conv2d(4, 4, kernel_size=(3, 3), stride=(1, 1), padding=(2, 2), dilation=(2, 2), bias=False)
(Incption2_conv4_1): Conv2d(11, 2, kernel_size=(1, 1), stride=(1, 1), bias=False)
(Incption2_conv4_2): Conv2d(2, 2, kernel_size=(3, 3), stride=(1, 1), padding=(3, 3), dilation=(3, 3), bias=False)
(Incption2_bn): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv1): Conv2d(12, 16, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(relu): ReLU(inplace=True)
(leakyrelu): LeakyReLU(negative_slope=0.1)
(bn1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(maxpooling1): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), padding=0, dilation=1, ceil_mode=False)
(conv2): Conv2d(16, 16, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), bias=False)
(maxpooling2): MaxPool2d(kernel_size=(4, 4), stride=(4, 4), padding=0, dilation=1, ceil_mode=False)
(dropout): Dropout2d(p=0.5, inplace=False)
(fc1): Linear(in_features=1024, out_features=16, bias=True)
(fc2): Linear(in_features=16, out_features=2, bias=True)
)