Espozo wrote:
So you're saying, for example, it can see that the pipe looks like a cylinder in that it's bright in the center and darker along the edges (and the pattern repeats vertically)? What I really want to know is how well this would work on an NES game with non perfectly front facing floors, (if that makes sense) like Battletoads. In order for this thing to work flawlessly (or even close to it) without hacks, it would practically have to be a human looking at the game. And just thinking, in the case of Battletoads, I guess it would also have to infer what is supposed to be up, and what is supposed to be further in the background for the characters.
Well we can say that the emulator is still in a very alpha stage. What the emu has succeeded to establish is the infrastructure, the framework. In the future, more semantic info will be collected from 2D scene,more predefined patterns will be added, more rules will be created to make the emu more robust. You say it right there will never be a perfect solution for this kind of problem but better solution is always possible. For the camera mode/direction detection: top down, left right or front back, perspective or orthographic or 2d projection, currently the emu has not detected it yet so the render will not be good for 3d camera or complex background game.
And when the emu becomes mature enough, the optimization for a specific game is as simple as values assigning for a set of coefficient. That's how i see things.