NeuroVis

The hardest part of learning CNNs isn't the math — it's not being able to picture what's actually happening. What exactly happens when the kernel slides across? Why does the feature map look like that? Why does adding one more layer help? Textbooks answer with symbols. NeuroVis shows you. Draw a digit with your finger, and a real CNN classifies it layer by layer in front of you — input tensor, convolution feature maps, activation, max pooling, and the softmax probability for each of the ten digits. Every stroke recomputes the entire chain instantly. Tap any feature map and the kernel sweeps across the input cell by cell, so you can see exactly where each output number comes from. Step through it, pause it, or just read the values off the grid. Wondering what those numbers inside the kernel actually do? Change one and watch the feature map follow. Wondering whether another layer matters?…