Visualization Tool

Overview

Netron is a neural network model visualization tool developed based on the Electron platform. It supports the visualization of many mainstream AI framework models and supports multiple platforms (such as Mac, Windows, and Linux). Netron supports MindSpore Lite models, allowing you to easily view model information. As shown in the following figure, after the .ms model is loaded using Netron, the topology structure, diagram, and node information of the model are displayed.

img

Functions

  • Load the .ms models. The MindSpore version must be 1.2.0 or later.

  • Display subgraphs.

  • Display the topology structure and data flow shape.

  • Display the format, input, and output of a model.

  • Display the type, name, attribute, input, and output of a node.

  • Display and save structured data such as weight and bias.

  • Export the visualization result as an image.

Usage

The code that supports the MS model has been merged into the official repository. The Netron is downloaded from https://github.com/lutzroeder/netron/releases/latest. The author may update the releases from time to time. After Netron is installed as follows, you can drag a model to the window to open it.

  • macOS: Download the .dmg file or run the brew cask install netron command.

  • Linux: Download the .AppImage file or run the snap install netron command.

  • Windows: Download the .exe file or run the winget install netron command.

  • Python server: Run the pip install netron command to install Netron, and then run the netron [FILE] or netron.start('[FILE]') command to load a model.

  • Browser: Open https://netron.app/.

Development and Debugging

Using the Development Version

Step 1: Use git clone https://github.com/lutzroeder/netron to clone and obtain a copy of the source code.

Step 2: Go to the netron directory and run the npm install command.

Step 3: Run the make build command for build and generate an executable program in the ./dist directory.

Debugging a Model Using Javascript

When debugging a model, add the information about the model to be debugged to the ./test/models.json file in the netron folder, and then use the node.js file to debug the ./test/model.js script.