Resolution Estimation for Shadow Mapping

We present an approach to efficiently reduce shadow map resolution while retaining high quality hard shadows. In the first step, we generate a list of sample points that are seen from the camera and then project these into light space, much like Alias-free Shadow Maps. In the next step, we analyze the list of sample points on the GPU to construct a tight light frustum for shadow rendering. After the light frustum is computed, we calculate for each sample the actual coverage in the final shadow map to estimate how large a shadow map pixel should be. From this number, we derive the lowest possible resolution to use in the shadow map while retaining nearly alias-free shadows. Our algorithm is built for a deferred renderer.