wtf is nn-svg?

NN-SVG is a browser-based tool for drawing diagrams of neural network architectures. Researchers working on machine learning papers often need these kinds of diagrams to illustrate how their models are structured, and drawing them by hand is slow and inconsistent. This tool lets you create them by setting parameters like the number of layers, sizes, colors, and spacing, rather than placing shapes manually. The tool supports three styles of diagram. The first is a classic fully-connected network diagram where each node in one layer connects to every node in the next. The second is a convolutional network diagram showing stacked rectangular feature maps, in the style used in early influential image-recognition papers. The third is a deeper, more modern-looking 3D-style diagram inspired by how architectures like AlexNet were visualised in academic publications. Once you have designed your diagram, you can export it as an SVG file. SVG is a vector format, which means it stays sharp at any size and is suitable for inserting directly into papers, presentations, or web pages without quality loss. The project is a web application hosted at alexlenail.me/NN-SVG and is also published as a research tool in the Journal of Open Source Software. The code is open source under the MIT license. Documentation and contribution guidelines are available in the project's wiki on GitHub.