I didn't look at the code, but canvas only supports 3x3 transforms and you need 4x4 to do perspective: https://developer.mozilla.org/en-US/docs/Web/API/CanvasRende...

So if those cubes are textured, you'd need to subdivide the texture and approximate each with affine transforms. Here's an example of this hack (which was only necessary years ago before webgl existed): https://acko.net/blog/projective-texturing-with-canvas/