The Leap Motion Controller is built on a unique mathematical model that maximizes speed and precision, so the raw data does not appear as a point cloud. Instead, at its most basic level, the API data arrives in frames. Because of this specialization, there are no depth images calculated anywhere at all.
The "point clouds" that were featured in our early videos are visualizations from some of our debugging tools. Because they’re not temporally or spatially consistent, they are not usable as methods of interaction. They also wouldn’t work for non-hand type objects, which just end up looking like a jumbled mess.
That said, we're working on using the hardware in novel and exciting ways. We hope to make new data types available in future versions of our API.