Pixel Size formula Conv Neural Network

main.js

function calcOut(wh, padding, kernelSize, stride) {
  return ((wh + 2*padding - kernelSize ) / stride) + 1;
}

// for image file 32x32 pixels
console.log(calcOut(32, 1, 3, 1)); // conv1 = 32
console.log(calcOut(32, 1, 3, 1)); // conv2 = 32
console.log(calcOut(32, 0, 2, 2)); // maxpool1 = 16
console.log(16*16*32);             // flattening result = 8192
console.log("----------------");

// for image file 28x28 pixels
console.log(calcOut(28, 1, 3, 1)); // conv1 = 28
console.log(calcOut(28, 0, 2, 2)); // maxpool1 = 14
console.log(calcOut(14, 1, 3, 1)); // conv2 = 14
console.log(calcOut(14, 0, 2, 2)); // maxpool2 = 7
console.log(7*7*64);               // flattening result = 3136
console.log("----------------");

// for image file 28x28 pixels
console.log(calcOut(28, 0, 5, 1));
console.log(calcOut(24, 0, 2, 2));
console.log(calcOut(12, 0, 5, 1));
console.log(calcOut(8, 0, 2, 2));

console.log("----------------");

// for image file 28x28 pixels
console.log(calcOut(28, 0, 5, 1)); // conv1 = 24
console.log(calcOut(24, 0, 2, 2)); // maxpool1 = 12
console.log(calcOut(12, 0, 5, 1)); // conv2 = 8
console.log(calcOut(8, 0, 2, 2)); // maxpool2 = 4

// 64 = batch size
// 20 = the number of channels
// 4 = width and height of the last output from maxpool2
console.log(64*4*4*20);   // 20480
console.log(20480/64);    // 320
console.log("----------------");