It is my belief that our intelligence intially comes from preprogrammed systems that are stacked on top of each other. To me, the path to human artificial intelligence lies in creating those building blocks, that will give us the power to manipulate more abstract concepts. Each one of these modules is supposed to be usable in any context, be it real life or video games.
Avaliable modules:
- Depth Map: Generates a 2D depth texture from a single image.
- Video Depth Map: Generates a 2D depth texture from a sequence of images.
Future modules:
- Layers estimator: Conditioned by depth, estimates each layer in a scene.
- Object estimator: Contitioned by 2D position, estimates the voxel contours and shape of an object.
- Movement Map: Generates a 2D map showing each pixel movement as a vector.
- Physics Estimator: Generates a 2D map of the physical properties of each pixel in an image.
- Scene Estimator: Uses multiple modules to generate a 3D voxel map with advanced properties.
- Scene Predictor: Uses a 3D voxel map to predict the future state of a scene (auto-regressive model).