Pytorch的網(wǎng)絡(luò)結(jié)構(gòu)可視化:Netron與TensorBoardX

• Pytorch的網(wǎng)絡(luò)結(jié)構(gòu)可視化:Netron
• Pytorch的網(wǎng)絡(luò)結(jié)構(gòu)可視化:TensorBoardX
• • 1.TensorBoardX簡(jiǎn)介
  • 2. tensorboardX的使用
• TensorBoard is not found.
• 參考資料

Pytorch的網(wǎng)絡(luò)結(jié)構(gòu)可視化:Netron

最近剛剛發(fā)現(xiàn)一個(gè)非常好用的顯示模型神器Netron

https://github.com/lutzroeder/Netron

https://www.electronjs.org/apps/netron

借助這個(gè)工具可以像windows的軟件一樣導(dǎo)入已經(jīng)訓(xùn)練好的模型加權(quán)重即可一鍵生成

以下是我的一個(gè)模型使用該工具可視化結(jié)果，只不過(guò)目前該工具對(duì)于onnx支持非常好，但是pytorch權(quán)重轉(zhuǎn)變?yōu)閛nnx是非常方便的,只需要torch.onnx.export()命令即可導(dǎo)出onnx權(quán)重

以下代碼將resnet18-5c106cde.pth轉(zhuǎn)化為resnet18.onnx

# -- coding: utf-8 -- import io import torch import torch.onnx import torchvisiondevice = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")def pth2onnx():model = torchvision.models.resnet18(pretrained=False)# pth_file = 'resnet18-5c106cde.pth'# model.load_state_dict(torch.load(pth_file, map_location=device))model.to(device)# data type n*c*h*wdummy_input = torch.randn(1, 3, 256, 256)dummy_input = dummy_input.to(device)input_names = ['actual_input']output_names = ['output']torch.onnx.export(model, dummy_input, 'resnet18.onnx', verbose=True, input_names=input_names,output_names=output_names)if __name__ == '__main__':pth2onnx()

resnet18網(wǎng)絡(luò)結(jié)構(gòu)圖

我目前看了下visdom實(shí)現(xiàn)pytorch的網(wǎng)絡(luò)結(jié)構(gòu)查找還是很困難，在stackflow上有很多人使用自己編寫的基于matplotlib來(lái)實(shí)現(xiàn)網(wǎng)絡(luò)結(jié)構(gòu)可視化適用性也不是很好，后來(lái)查找到使用基于tensorboard所開發(fā)的tensorboardX可以很方便的實(shí)現(xiàn)pytorch網(wǎng)絡(luò)結(jié)構(gòu)的可視化，因此決定采用這種方式。

Pytorch的網(wǎng)絡(luò)結(jié)構(gòu)可視化:TensorBoardX

1.TensorBoardX簡(jiǎn)介

tensorboardX的項(xiàng)目路徑：https://github.com/lanpa/tensorboardX

tensorboardX是基于tensorboard的思想用來(lái)寫tensorboard events的工具，可以實(shí)現(xiàn)對(duì)傳統(tǒng)的tensorboard中 scalar，image，figure，histogram，audio，text，graph，onnx_graph等事件進(jìn)行編寫。

tensorboardX同時(shí)具有論壇供大家提出問(wèn)題解決問(wèn)題 ，論壇地址：https://github.com/lanpa/tensorboardX/wiki

2. tensorboardX的使用

tensorboardX的安裝以及依賴如下所示：

pip install tensorboardpip install tensorflowpip install tensorboardX

tensorboardX的路徑下帶的有一個(gè)規(guī)范的demo，可以供大家參考。我這里公布一個(gè)我自己測(cè)試過(guò)的代碼，代碼來(lái)源于：https://blog.csdn.net/sunqiande88/article/details/80155925

# -- coding: utf-8 -- import torch import torch.nn as nn from tensorboardX import SummaryWriterclass LeNet(nn.Module):def __init__(self):super(LeNet, self).__init__()self.conv1 = nn.Sequential( # input size=(1*28*28)nn.Conv2d(in_channels=1, out_channels=6, kernel_size=5, stride=1, padding=2),nn.ReLU(), # (6*28*28)nn.MaxPool2d(kernel_size=2, stride=2), # output_size(6*14*14))self.conv2 = nn.Sequential(nn.Conv2d(in_channels=6, out_channels=16, kernel_size=5),nn.ReLU(), # (16*10*10)nn.MaxPool2d(kernel_size=2, stride=2) # output_size=(16*5*5))self.fc1 = nn.Sequential(nn.Linear(in_features=16 * 5 * 5, out_features=120),nn.ReLU())self.fc2 = nn.Sequential(nn.Linear(in_features=120, out_features=84),nn.ReLU())self.fc3 = nn.Linear(84, 10)def forward(self, x):x = self.conv1(x)x = self.conv2(x)# Linear的輸入和輸出都是維度為1的值，所以要把多維度的tensor展平成一維x = x.view(x.size()[0], -1)x = self.fc1(x)x = self.fc2(x)x = self.fc3(x)return xdummy_input = torch.rand(4, 1, 28, 28) # 假設(shè)輸入4張1*28*28的圖片 model = LeNet() with SummaryWriter(comment='LeNet') as w:w.add_graph(model, (dummy_input,))

運(yùn)行該代碼后會(huì)自動(dòng)生成一個(gè)runs文件夾，并且在文件夾下會(huì)有一個(gè)對(duì)應(yīng)的event，如下圖所示：

此時(shí)需要在terminal或者cmd下運(yùn)行tensorboard --logdir = path
此處千萬(wàn)要注意，如果按照上面的參考文檔的方式是會(huì)報(bào)錯(cuò)的：No graph definition files were found 或者 No definition files were found，總之無(wú)法正常顯示網(wǎng)絡(luò)結(jié)構(gòu)圖。

此處的path 是event對(duì)應(yīng)的確切，完整的路徑

在運(yùn)行后會(huì)出現(xiàn)一個(gè)http url，此時(shí)需要將該url 拷貝到chrome下即可看到如下所示框圖

結(jié)構(gòu)框圖如下所示：

此時(shí)雙擊紅圈所示的LeNet模塊即可看到LeNet的細(xì)節(jié)信息，如下所示：

至此即可完成使用tensorboardX 對(duì)pytorch網(wǎng)絡(luò)結(jié)構(gòu)的可視化

TensorBoard is not found.

Just started using Tensorflow, but I am not able to use tensorboard command on my cmd, it gives the error command

C:\Users\tushar\PycharmProjects>tensorboard --logdir="NewTF" 'tensorboard' is not recognized as an internal or external command,operable program or batch file.

I had the same problem for tensorflow 1.5.0 and windows10.

Following tensor documentation (“Launching TensorBoard” section), you can try:

python -m tensorboard.main --logdir=[PATH_TO_LOGDIR]

參考資料

https://blog.csdn.net/xiaoxifei/article/details/82735355

https://blog.csdn.net/lei19880402/article/details/103721362

總結(jié)

以上是生活随笔為你收集整理的Pytorch的网络结构可视化:Netron与TensorBoardX的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。