Possible speed up with LOD for proxies, scatter and shaders?

[lupaz]

Reading on Stack (Facebook), I found this comment. Could this be implemented in Corona? (see attached)
Especially useful, I think, for Scatter objects.

[maru]

Yes, this is something we are actually investigating.

I am not saying that using different LODs does not ever make sense, but on the other hand: if you instance X high-poly objects, the RAM usage will be actually lower than if you instance the same number of high-poly objects plus low-poly objects. :)
This is because in the first case only one model is loaded into memory and in the second case there are two different objects.

[lupaz]

Yes, this is something we are actually investigating.

I am not saying that using different LODs does not ever make sense, but on the other hand: if you instance X high-poly objects, the RAM usage will be actually lower than if you instance the same number of high-poly objects plus low-poly objects. :)
This is because in the first case only one model is loaded into memory and in the second case there are two different objects.

Got it. Glad you're investigating it.

i'm particularly interested in the "not rendering occluded scatter instances" part. I can imagine it making a lot of sense when rendering a house for example, in which the lawn scatter object is surrounding it. So with each camera the house blocks a huge part of the lawn scatter object.

[romullus]

Some native LOD system for the proxies and the scatter would be very appreciated indeed. I think that slight increase in RAM usage would be greatly offset by increased overall performance.

[maru]

Things like this always have the potential of making some scattered objects invisible, for example in reflections, or having shadows of instances visibly missing, etc. But yes, I can confirm that we are investigating LODs, especially for Scatter.