An interactive book
Receptive Field.
Convolutional neural networks, one idea at a time.
Every chapter is built around something you can drag, slide or step through, so the shapes and numbers stop being things you memorize. Short modules, each ending in a knowledge check, take you from a single kernel to real architectures.
Table of Contents.
- 01Why not a dense layer?soon
Flatten an image and watch the parameter count explode.
- 02Convolution by handsoon
Slide a kernel and compute each output as a dot product.
- 03Stride, padding and dilation4 modules
Three settings that decide where the kernel lands and how big the output is.
- 04Channelssoon
Why kernels are really three-dimensional.
- 05Pooling4 modules
Max and average pooling, and which input wins.
- 06Receptive fieldssoon
Stack convolutions and pooling layers, and watch how much of the image one output can see.
- 07Backprop through a convolutionsoon
Where the gradients go.
- 08Upsampling and transposed convolutionsoon
Spreading values back out, and checkerboard artifacts.
- 09What the network learnedsoon
A real CNN running in the browser, layer by layer.
- 10Architectures as storiessoon
LeNet to ResNet to U-Net, one idea at a time.