Matthew Vitelli provided me with another approach to the transmittance technique.
What he is doing is to compute the transmission specially for each object. This would allow you to have specialized Du and C components for each object and also reduces the memory usage on the GPU, as well as the pixel overdraw which comes from computing transmission as a fullscreen quad. He has modified the 2-layer transmission approach to include these features. Now, only three render-targets are created in total (1 for near-back face depth, the other 2 for far depth) and each are of half-single (16-bit) precision. The position reconstruction technique works excellently with this. Also, the benefit of doing transmission outside of a post-process lets you have specialized parameters for each object.
What he is doing is to compute the transmission specially for each object. This would allow you to have specialized Du and C components for each object and also reduces the memory usage on the GPU, as well as the pixel overdraw which comes from computing transmission as a fullscreen quad. He has modified the 2-layer transmission approach to include these features. Now, only three render-targets are created in total (1 for near-back face depth, the other 2 for far depth) and each are of half-single (16-bit) precision. The position reconstruction technique works excellently with this. Also, the benefit of doing transmission outside of a post-process lets you have specialized parameters for each object.
Matthew Vitellis technice can be found here: Executable + Source
Thanks again for shareing this with me, and the readers of my blog!
Modern art using the GPU 