Basically, what it does is look at half of a symmetrical 12*8 image and tell you if that image is a sprite of a space invader or not.
So far, testing is apparently successful, with the total “incorectness” lowering continually. The desired outcome is that the space invader sprites give a “result” of 1 and the non space invader ones give 0. Testing is slow, but you can close and continue testing where it left off, without fear of data loss.
You might ask, why would this be useful at all? Well,
The plan is to build a second neural network that generates images from random noise. The first neural net will be tasked with detecting if the image was generated, or if it is an original space invader. We will train the neural networks until the second generates good looking space invaders, and then voila, we have a way of generating space invaders easily and quickly.
Get LowRes Coder to use this program.