Netron, a program to visualize neural network models

about Netron

In the next article we are going to take a look at Netron. This is a program to view models of neural networks. This application that uses Electron / NodeJS and is published under the MIT license, we can run it on Gnu / Linux, macOS, Windows systems and from the web browser.

This program was created by Lutz Roeder. Netron is an open source tool that allows you to visualize neural network models, which also it will allow us to analyze the structure of the model and thus ensure that it matches the expected design. It is software compatible with a variety of frameworks and model formats.

Netron supported formats

Netron has support for formats as they are:

  • ONNX (.onnx, .pb, .pbtxt)
  • Keras (.h5, .keras)
  • TensorFlow Lite (.tflite)
  • Caffe (.caffemodel, .prototxt)
  • Darknet (.cfg)
  • Core ML (.mlmodel)
  • MNN (.mnn)
  • MXNet (.model, -symbol.json)
  • ncnn (.param)
  • PaddlePaddle (.zip, __model__)
  • Caffe2 (predict_net.pb)
  • Barracuda (.nn)
  • Tengine (.tmfile)
  • TNN (.tnnproto)
  • RKNN (.rknn)
  • MindSpore Lite (.ms)
  • UFF (.uff)

netron running from desktop

In addition Netron also has experimental support for; TensorFlow (.pb, .meta, .pbtxt, .ckpt, .index), PyTorch (.pt, .pth), TorchScript (.pt, .pth), OpenVINO (.xml), Torch (.t7), Arm NN (.armnn), BigDL (.bigdl, .model), Chainer (.npz, .h5), CNTK (.model, .cntk), Deeplearning4j (.zip), MediaPipe (.pbtxt), ML.NET (.zip ), scikit-learn (.pkl), TensorFlow.js (model.json, .pb).

Install Netron Neural Network Viewer on Ubuntu

Test from the web browser

netron running from web browser

Before deciding to install this program, we can choose to test it from web browser. If you don’t have a model that you can upload to test it, you can use the sample model examples that can be found in the repository on GitHub of the project, to download or open with this browser version.

As snap package

If you decide to install this software on your computer, You can install this program through its snap package, which can be found at Snapcraft.

As I said, Netron Neural Network Viewer can be installed on Ubuntu via Snap by doing the following. To start we will need to open a terminal (Ctrl + Alt + T) and then we will install the stable version of the program using the command:

install netron as snap

sudo snap install netron

After installation, in case you need update the program, in a terminal you just have to execute:

sudo snap refresh netron

After all the above, we can start the program from the Applications menu or from any other launcher that we have available in our distribution. In addition, we can also start it by typing in the terminal (Ctrl + Alt + T):

netron launcher

netron

Uninstall

For uninstall Netron Neural Network Viewer installed via Snap package, we will only have to execute in a terminal (Ctrl + Alt + T) the command:

uninstall Netron snap

sudo snap remove netron

Download AppImage

We can also use this program using the AppImage package that can be downloaded from the project release page. In addition to being able to download this package from the web browser, we will also have the possibility of using wget to get hold of the file.

For download the latest version published today, we will only have to open a terminal (Ctrl + Alt + T) and execute in it:

download netron appimage

wget https://github.com/lutzroeder/netron/releases/download/v5.3.4/Netron-5.3.4.AppImage

When the download is finished, we have give execute permissions to the file that we just downloaded. For this, if we move to the folder in which we have the file saved, we will only have to execute this command:

sudo chmod +x Netron-5.3.4.AppImage

After the previous command, we can start the program by double clicking on the file, or by typing in the same terminal:

install netron as appimage

./Netron-5.3.4.AppImage

Netron is a simple way to visualize neural networks. This program will allow us to use a wide range of frames and compatible model types. It is really scalable and usable for many people in the learning community. Graphics can even be exported, although you may want to use a different approach if your goal is to generate graphics for printing, especially when they are very deep.

Users who want to, can get more information about this program at project website or in your GitHub repository.

Add Comment