Netron

Netron is a tool designed for visualizing machine learning models. It supports multiple model formats and provides a graphical representation of their structure. It is compatible with frameworks such as TensorFlow, ONNX, Core ML, and others.

★★★★★

5.0(1 ratings)

• Model visualization. Netron displays the architecture of machine learning models as a graph. Each layer and operation is represented as a node, making it easier to analyze the model's structure.
• Multiple format support. The tool supports formats such as .pb (TensorFlow), .onnx (ONNX), .h5 (Keras), .mlmodel (Core ML), and others. This lets you load models trained in different frameworks.
• Parameter exploration. Netron lets you inspect the parameters of each layer, such as weights, biases, and configuration attributes. Values are shown in an organized interface.
• Information export. Model details, including layer names, dimensions, and operation types, can be exported for further analysis, along with an image of the graph in PNG or SVG format.
• Interactive navigation. The model graph can be zoomed in, zoomed out, and panned to examine specific sections. Nodes can be selected to view additional details.
• Quantized model support. Netron identifies and displays information about quantized models, including scales and zero points in quantized operations.
• Metadata compatibility. If the model includes metadata, such as layer descriptions or versions, Netron shows it in a dedicated section.
• Offline mode. The tool works without an Internet connection, which ensures privacy and access in restricted environments.
• Command-line integration. Netron can be launched from the terminal, which allows it to be used in automated workflows.
• Cross-platform interface. Available for Windows, macOS, and Linux, as well as a web version that requires no installation.

Netron was created by Lutz Roeder, and its development began in 2017. The project is written in JavaScript and uses Electron for the desktop version. The web version is based on standard technologies such as HTML5 and CSS. The source code is hosted on GitHub under a permissive license, which allows its use and modification. Over the years, Netron has added support for new model formats and visualization improvements. The development community contributes bug reports and suggestions for new features.

Alternatives to Netron: