Making CNNs legible for the rest of us

What it does — CNN Explainer is an interactive visualization system that runs in the browser to teach non-experts how Convolutional Neural Networks work. It visualizes the inner workings of a live Tiny VGG model, showing how input data is processed layer by layer. The whole thing is accessible via a public URL without installation, though the source is available if you want to run it locally.

The interesting bit — The project is backed by a 2020 IEEE TVCG paper, which means the visualization choices—what to show, what to hide, and when to reveal it—were research-grade decisions aimed at pedagogy rather than eye candy. That academic rigor shows in the restraint: it explains one small model deeply instead of letting you drop in any arbitrary architecture.

Key highlights

• Designed for non-experts rather than ML engineers tuning hyperparameters.
• Ships with a fully trained Tiny VGG model and its training pipeline in ./tiny-vgg/.
• Runs entirely in the browser via a live demo, with no backend required.
• Published in IEEE TVCG (2020) by a Georgia Tech–Oregon State collaboration.
• Extending it to custom CNN models or image classes requires consulting specific GitHub issues (#8 and #14); the README does not document a plug-and-play API.

Caveats

• Adapting the tool to your own CNN model or image classes is possible, but the README offers no formal extension guide and instead routes you to GitHub issues #8 and #14.

Verdict — Ideal for educators, students, or anyone who understands CNN theory but still cannot picture what a kernel actually does. If you already train custom models and need a general-purpose architecture debugger, this is not your tool.