Thoughts and implementations on VR (mainly in virtual city scenarios) (may be serialized)

This is also my graduation project. I was thinking about writing this article for a while. It seems that the time is not enough, so I have to hurry up.

I used to focus on games. I think there are two main outdoor scenarios: terrain and staticmesh. The former is difficult to use in urban scenarios. The latter has high requirements on art, heavy workload, and difficult to modify. It should be said that we do not know anything about VR. We feel that the current application is not very wide and will only be used in some large projects or scientific research, because of the cost.

The recently written item is the dynamic mesh, Which is dynamically generated by ctrlpoint and brush. Ctrlpoint is three-dimensional coordinates, which are used to connect to a curve. A brush is a variety of brushes that encapsulate different algorithms and shader and then generate a mesh along the curve. It sounds very simple, but it is actually very simple :)

Does dynamic mesh have a large amount of computing? It should be said that it will not be larger than the skeleton mesh, and a considerable amount of computing can be placed in. For some commercial machines with powerful CPUs, using softvs does not affect performance, and pixel processing of fixed pipelines can also meet basic applications.

What is the efficiency of dbuffer? In fact, as long as the previous slave has been painted during the lock, the CPU will not stop. An intuitive way is to prepare two groups of buffer and fill the other group when drawing one group.

In some cases, you cannot use staticmesh, but instancing of staticmesh is a good thing. You can also use a brush to calculate the position and direction of these instances.

The global quad-tree stores only the curves of ctrlpoint and their connection. Each shard only needs to generate visible mesh. With limited memory, the scenario can be quite large, there is no dynamic loading. When constructing a visible list, roughly follow the order from near to far. If the number of instances exceeds the maximum value, draw the nearest n instances, which will not significantly affect the image.

The current progress of this item is still very low, and its availability is still unknown. In any case, it is a pleasure to implement your own ideas, if you are interested in this item, please let me know ^!