Biological neurons function in a very different way, as compared to the simplistic artificial neural networks of machine learning. For example, see how real neurons work and how they connect with each other. The types of neurons themselves are very varied: "...neurons to take specialized forms such as unipolar,bipolar, multipolar, anaxonic, pseudounipolar, basket cells, purkinje cells, Lugaro cells, spindle cells and more.".
As for the eyes, light passes through a large network of neurons before hitting the rod and cone cells. It's the opposite for the octopus retina. Animal babies are able to learn to walk much quicker than human babies. Also, have a look at videos of how blind children learn to walk.
So essentially, there's a lot lot lot more going on in the creation and working of the human body than just "neural network training". A really nice paper titled "facts and anomalies" gives perspective into this complexity.
Nobody fully understands how the system of cells and the equilibrium works. It's extremely complex and multi-dimensional. When you look at the complexity inside a single cell and the electron transport chain, you'll notice that there's a heck of a lot of things that are already pre-programmed and designed to make use of the molecular properties of nature to create structures and motors using the equilibrium of chemical properties. When such an equilibrium is created, evolution and adaptation can happen, but only in a very controlled manner. So I do not believe a baby's neural network is getting trained to walk or see. It's pre-trained from millions of years of evolution. The reason it may be taking some time to mature, is probably because the corresponding neurons in the brain are taking some time to get created, grow and find their way in the brain to form the right synapses.