NN-SVG: a parametric generator for publication-ready neural network architecture schematics

What it solves

NN-SVG solves the tedious and time-consuming process of manually drawing neural network architecture diagrams for academic papers and web pages. Instead of constructing these diagrams from scratch by hand, researchers can generate them parametrically.

How it works

The tool uses JavaScript libraries to programmatically generate figures based on user-defined parameters for size, color, and layout. It supports three specific styles of architecture drawings:

• Fully-Connected Neural Networks (FCNN): Generated using the D3.js library.
• Convolutional Neural Networks (CNN): Based on the LeNet style, generated using D3.js.
• Convolutional Neural Networks (Deep): Based on the AlexNet style, generated using Three.js for 3D-like representations.

Who it’s for

Machine learning researchers who need publication-ready schematics of their model architectures, as well as educators who may use the tool for pedagogical purposes.

Highlights

• Parametric Generation: Create diagrams without manual drawing.
• SVG Export: Export figures as Scalable Vector Graphics for high-quality inclusion in papers.
• Multiple Styles: Support for FCNN, LeNet-style CNNs, and AlexNet-style deep networks.
• Customizable Styling: Control over size, color, and layout parameters.