Flexible Termination of LC pass

[dalomar_cg]

I'm not sure if this is a weird request or not, but I think it would be great to be able to have flexibility in how long the LC pass takes to calculate. An example of this would be that maybe the LC subdivs get set for 3000, but somewhere around 1800 subdiv you realize that the amount of sampling is enough for your desired quality. It would be good to be able to simply kill the rest of the LC calculation, but not the rest of the render. This is similar to ppt's ability to be terminated at any point in the render, but simply with LC light pass (and have it pass on to the true "rendering").

[Micha_cg]

Re: Flexible Termination of LC pass

Great idea. The cancel button could be sinsitive for what kind of calculation is running.

If cancel during LC, than start next calculation, else complete stop.

[dalomar_cg]

Re: Flexible Termination of LC pass

Maybe it could even work during the IR pass...either canceling would prevent further prepasses (ie 4 prepasses are set, but canceling midway through the 3rd pass will let the 3rd pass render and go straight to the rendering), or even let the current buckets finish and go on to rendering. (same example, but canceling prevents any new buckets from calculating the 3rd pass which would lead to certain areas only having 2 prepasses and others having 3)

[Micha_cg]

Re: Flexible Termination of LC pass

I tested multipass vs. single pass and use single pass only now.

Hey, if you extend this method of stop-and-go, than at the end you could come to the progressive QMC sampling.
I hope Vlado don't forget it and implement it.

[dalomar_cg]

Re: Flexible Termination of LC pass

In my tests multipass has always been faster than single pass. Singlepass also tends to need more time to calculate the sampling at the edge of the buckets which leads to longer times as well as bucket artifacts (depending on other aspects of course).

I know what your thinking of this leading towards your pqmc method, but that wasn't the intent. It was more to make sure that you could set a very high subdivision count for lc yet not still kill it half way if were satisfied with the quality of the LC pass not the image as a whole. I've got some comments about the pqmc method, so I'll move those over to that thread.