I see that convolutional neural networks are used specifically for image recognition and object detection since their hidden layers and nodes are designed to encode features such as 3 dimensionality, shadows, depth, and other common things found in real life images.
My specific interest is computer vision for desktop computer applications and computer games. Screenshots from a computer don't contain all of the same features as real life images, but nevertheless there must be some overlap particularly from games. I'm curious whether convolutional neural networks would be useful for this type of artificial object recognition, or if there is a more appropriate type of neural network architecture? Also would it make a difference in the object detection in computer games whether those games were 2D, 2.5D or 3D in terms of which architecture is best?