Cascaded shadow map

Recently, a new real-time shadow rendering technology is used to render large scenarios. It seems to be very effective, but one drawback is that the scenario needs to be rendered multiple times, its central idea is to scatter screen pixels to multiple shadowmaps. Because the details of near-side objects are high and the details of distant objects are low, they are separated by the vertebral body, and focus the light source camera on the corresponding small cone. Of course, there is a mathematical method to derive the separation of the cone. For details, see the PSSM or light Space Perspective shadow maps thesis, there is a mathematical derivation process above, and the shadow acne will be generated in the shadow Paster process, which is caused by the accuracy of Z buffer. It is generally corrected using Z bias and slope bias, today, gamedev saw an empirical formula using 1.0f/(d * D-2d) as bias. D represents the distance from the light source to the vertex. I tried it, I don't feel much useful. This can also be improved using the back face render method. The effect is good, but for some Some problems may occur on the back of continuous objects (such as terrain), or the Z-precision problem. For some objects with large variation, such as houses, the results are still very good, if variance shadow ing is added, we can get a very delicate soft shadow. I am not very clear about this principle. I am not quite familiar with mathematical statistics. PSSM + VSM seems to be the technology used by cryengine2.