camera path mode

[RE:FORM_STUDIOS]

I'm noticing that to retain decent quality in my GI when using camera path mode with IR and LC (prepass / animation) for rendering scenes with moving objects, I need to push my GI and LC sampling rates way up. The shorter the sequence the lower the GI I can get away with and the better the result.

Is there some kind of sweetspot, maybe at just 50-100 frames or so where you can use pretty low GI settings without losing quality? If so, instead of manually breaking up an animation into segments of this duration and then re-stitching them together in post, would it be worthwhile adding an option in the GI settings to split the camera-path sampling into seperate chunks of time ?

or... am I completely mis-understanding how this all works

[Da_elf]

i didnt know you could use camera path with moving objects

[duke2]

I'd love to hear more information on this too. Without hijacking the thread, i'd also like to know how many samples you should be feeding it? If it's 100 frames long, in a scene that used to be incremental add, every 10 frames, should you just (HSPH * frames) / 10 ?

[RE:FORM_STUDIOS]

I'm up to my neck with a deadline looming so to save time trying to find a sweetspot for my GI settings to render a 125 or 250 frame sequence, I'm breaking my GI prepasses down into 25 frame sequences... fingers crossed there's not a jump in the GI slolution when I render the final 125 frame pass... !

Initial results are good and fast, but I've not tried rendering over two prepass 'blocks'

Hope vlado can confirm whether my approach holds any advantage, or whether I should be able to achieve similar quality of GI without breaking the prepass down into shorter segments.

[RE:FORM_STUDIOS]

Ok, the deadline passed, and my settings seemed to work OK... but not perfect.

I am getting some strange artifacts using camera path dispite using fairly hi GI settings.
Does anyone know what might be causing this effect? http://dl.getdropbox.com/u/1620039/C05B_R_s1_2.mov

Prepass Render Settings are :-
Adaptive DMC 1/9 Clrd Thresh 0.015 Use DMC thresh unticked
IR prepass mode, use camera path checked
IR min max -2/-1
IR Hsph 75
Clr Thresh 0.3
Nrm Thresh 0.1
Dist thres 0.1
LC Single frame mode, use camera path checked
LC subdivs 1422
LC Sample size 0.02
LC scale screen
LC Prefilter 50
LC Filter none

Final Render Settings are :
Adaptive DMC 2/7 Colour thresh 0.016
IR rendering mode, use camer path checked
IR settings same as above but Interp samples 23, Interp frames 2
LC off.
DMC Sampler Adaptive = 0.85, Noise Thresh 0.008, min samples 8, subdiv multiplier 1.0

Render times for 720 x 405 frames were around 10 mins for prepass 15 mins for render on a i7 clocked at 3.6Ghz. so no slouch!

Any ideas/suggestions?

// edit /// on further investigation, it appears this artifacting is caused by the LC interp samples being set too high, reducing them down removes the strange shapes swimming in the GI, but then you end up with blotchy GI. I was using "least squares fit" method, changing this to "delone triangulation" makes a huge improvement.

[werticus]

When using LC in single frame mode + camera path with moving objects and moving camera.
(and ir map with pre pass / final rendering mode)

Does the number of subdivisions need to be made very high like it does with fly through mode?

I see that its calculating the full set of subdivisions over the entire range for every frame so I would think not or the render times would blow out?

Its my understanding you dont need to use LC at all for the final IR pass though.

Be nice if there was some clarity on how to use this properly, the spot 3d tutorial is very light.