CNN Vision Lab

lab/11 · real CNN · feature maps · adversary

A real convolutional network, trained on MNIST and running in your browser. Draw a digit and watch every filter develop like a print in the darkroom: eight edge detectors, then sixteen shape detectors, then the meter swings to a guess. Turn on saliency to see which pixels mattered. Then play the adversary: flip a 98%-accurate model's mind with as few pixel edits as you can.

lab/11 · game

CNN Vision Lab

real CNN · feature maps · adversary

> warming phosphor

← LabLab blogHome