Exterior animation lightcache setup

[Noiré]

We have an exterior scene whose extents is at least 20.000 meters. The system units are set to meters. We thought of using lightcache for the secondary GI bounces, but have problems finidng the correct settings. Vlado recommends using screen scale samples for exterior fly-through LC's, and not world scale. Trying with a screen scale of 0.5m and a subdiv of 2500 for a 700 frames 768x440 sequence. The lightcache calculation seems to be endless. With a world scale of 0.02 meters on the other hand, the lightcahe lands at around 13MB and is not good enough.

Can anyone recomend some good settings for our setup? We have also tried QMC for secondary, and the result is fine, but render times end up at around 30 min a frame, which is a bit too much.

[alfienoakes]

Use this tutorial for rendering an animation. (thisone is for an animated camera)

 

Shows you about pre-calcing the IR maps and LC etc.. very good:

 

http://www.spot3d.com/vray/help/VRayHelp150beta/tutorials_imap2.htm

 

Andy

[Noiré]

yes, thank you, the tutorial you link to is good, but our setup is quite different, not being an interior shot for one, in addition to being a much larger scene, and having different units.

Maybe rephrasing my question can clarify: Would you recomend using lightcache for secondary bounces for a highpoly (750 onyxtrees) exterior scene with very large extents, and, in that case could you point me towards reasonable settings, or would you rather reccomend brute force secondaries for this type of scene?

[manta]

How many computers are you using for rendering ?

[Noiré]

Currently three. Actually, I believe I found a way to send a complete job to backburner (not having to manually intervene between the steps)following these 3 steps:

1. Send the lightcache calculation for the entire sequence as a single job assigned to one render-node, saving out the lightcache.

2. Send the primary bounces calculation (multiframe incremental IR-map, every 10-15 frames) using the (not yet ready) saved lightcache for secondary, making this job dependent on the 1st job, so it does not start until the lightcache is finished.

3. Send the final rendering, dependent on the 2nd job, using precalced lightcache and IR map.

Correct me if I'm wrong about these.

Concerning the lightcache settings though, I gave up waiting for ages trying to get them right, and went for QMC secondary bounces. I'd still be grateful though if anyone will share some insight on this issue.

[manta]

The reason I asked was if it takes 30 min. for a frame, then with three computers its only 10 min. per frame...is that still too long ?

[alfienoakes]

If your using Lightcache for your secondary bounce, on an external scene, and you use the "world" setting, you need to make sure all your geometry is fairly close together.

 

What I mean is, if you had an external scene that was say 10 km wide, then lightcache would take forever as there is so much "world" to calculate.

 

The units you use are important but the principles are the same, you just need to allow for the type you are using (metric, imperial.. etc).

 

Your subdivs are quite high at 2500, and your world sample size of .02 is the equivalent of a 20mm sample (if your scale of units = 1m)

 

But for screen, and I am willing to be corrected here) doesnt work the same. 0.5 sample size is small, very small. Thats not in metres. I think its based more on your screen size which is pixels. So you have a very small sample size. Try upping screen sample size to 50 and then decrease it.

 

The reason Vlado recommends screen on an external animation is becuase of what I said above, it depends what extent your "world" covers..

 

I may have confused myself there, so if anyone pulls me on on anything.. cheers..!

 

Andy

[Noiré]

Yes, I think the conclusion is that world scale is quite useless for larger scenes. Checking up on the help file regarding Screen scale, I quote "screen - the units are fractions of the final image (a value of 1.0 means the samples will be as large as the whole image)," which probably means that a good setting for screen scale is somewhere between 0.02 and 0.05.

Regarding the relation between the length of the sequence and the lightcache settings, I believe now that the subdivs are what matters here. This is the reason why the subdivs need to be significantly higher for a flythrough calculation compared to a single fram render. I guess the reason for the subdivs not needing to be 700 times greater (bear in mind that the figures are square roots of the actual number of paths traced, so 3000 sbdvs are 9x of 1000sbdvs) for a 700 f sequence is that one gets a lot of overlaps for the different frames. The speed the camera travels at probably matter quite a lot. I am still short of a good way of estimating the settings, but I have at least understood the lightcache a little better.

[alfienoakes]

Hey Erik,

 

We have all learnt something here...

 

Now for a solution. I have had a look through sites on exterior animations, and it seems that IR map / QMC is the preffered method. Its then a case of getting the settings working for each of these.

 

Its definitely worth running your IR map every n'th frame, and then on the final animation, using the pre calc'd ir map file, which I think you are doing anyway.

 

I would concentrate on the QMC settings, and getting those correct, as it seems this is the way to go. Unfortunately, Chaosgroup site is down for maintenance at the mo, so I cant even paste any links to their site on the settings..

 

This link at least tells you what each one does, you may have seen it:

 

http://www.spot3d.com/vray/help/VRayHelp150beta/render_params_qmc.htm

 

Andy