GI baking, UHDC restrictions and Animation

[FrostKiwi]

Hey,
I know this has been mentioned multiple times in same shape or form, yet would like to give my two cents aswell:
While making a low quality flythrough test of a Scene for an animated short, -> https://forum.corona-renderer.com/index.php/topic,6579.0.html
I had to optimize for minimal Passes count, so that I won't take more than 5 minutes per frame in the final render, thus I oversized normalmaps (fewer samples needed), no shadows on posters etc.
Then I turned off secondary bounces (almost no visual difference) and to my surprise it delivered a more noisy result in the same time span (no interpolation of primary rays I suppose, when no UHDC), so UHDC comes back on. I tested this multiple times and with UHDC it's always a cleaner result in less time (including prepass time) [minus the animation flickering obviously]

• Thus it would be nice to be able to bake Primary rays, just as secondary rays. (for flythroughs) (assuming it's possible with how refraction doesn't work without the first bounce)
  
• UHDC is limited to 200k bounces and is (semi)view bound. It would be nice not to have to make multiple UHDC files per room (and camera angle in larger rooms), I know there is the append function, but an hdc cache realtime view would be nice to see which angles have to get more gi samples). I know this goes against what corona stands for (less optimization and more artwork time) but rendertimes for flythrough are still a harsh reality and the benefit of baked gi is just too high imo.
  As a more surreal solution: maybe just show the raw GI pass and fly around in interactive mode, while it keeps adding sample where you look at, and showing the samples already cached.
  
• Is it possible to rip-of V-ray and include a cast rays along camera path, and unlock GI bounces count beyond 200k?

[maru]

I had to optimize for minimal Passes count, so that I won't take more than 5 minutes per frame

This doesn't make much sense. You could as well decrease gi/aa or lsm to have MORE passes in the same time to get better antialiasing.

in the final render, thus I oversized normalmaps (fewer samples needed)

What do you mean?

no shadows on posters etc.

What do you mean again?

and to my surprise it delivered a more noisy result in the same time span (no interpolation of primary rays I suppose, when no UHDC)

When you are only using path tracing, then this is pretty obvious. :)

UHDC is limited to 200k bounces and is (semi)view bound

What do you mean?

[FrostKiwi]

This doesn't make much sense. You could as well decrease gi/aa or lsm to have MORE passes in the same time to get better antialiasing.

Ye a bit unclear: To get a near clear frame with only 5min. per frame render time.

What do you mean?

Unneccecary sidenote of mine, I mean I make the normalmaps bigger than normal, so that the reflections aren't as detailed and won't need a ton of AA passes to look good in motion.
With no shadows on posters I mean I get rid of shadows on small details, you would normaly ignore ones motion is added, for instance pictures on the wall

When you are only using path tracing, then this is pretty obvious. :)

No it isn't! Why should having only 1 GI bounce be more noisy than 2...3...4... gi bounces?

What do you mean?

The problem with UHD cache is that when you put the camera for UHDC pass in one direction and get for instance 70% gi cache sucess, the other direction get's only 15%, because coverage is missing for that angle.
Also, once I reach 200k bounces on the GI cache, it always stops caching more bounces! Since the Box for Max GI samples from the HDCache is gone in the UHD Cache.

[Ondra]

you have flickering with UHD cache? can you report it as a bug?

Other than that, I really dont understand much of the topic. Some biased primary GI solution will be probably implemented in the future.

[FrostKiwi]

you have flickering with UHD cache? can you report it as a bug?

Other than that, I really dont understand much of the topic. Some biased primary GI solution will be probably implemented in the future.

Pretty sure it is. Check from 0:09 to 0:20 on the vid to confirm.
http://saires.de/rendertest

The whole second room is just one big 200kbounces pre-cached .hdc
(My theory is, some angles my have been missed by the precache, thus being retraced in ultra crap quality at rendertime)

[Ondra]

is that the latest build, default animation settings? If yes, then I would like to take a look at the scene (can you create a bug report)?

[FrostKiwi]

is that the latest build, default animation settings? If yes, then I would like to take a look at the scene (can you create a bug report)?

Bug report on mantis. I might have actually found the reason: flickering happend where a decal was placed nearby. I alligned the decals with the position of the walls and pushed their postition one tick away from the wall. Even though they are infront, they still flicker in the viewport, but do not in the on the final render. Maybe the GI calc has Z-depth issues? (Embree precision and high quality were on, I didnt play with it back then)

build was 13.01.15
I made periodic saves, so this is the file that was used back then.
it's actually 7k objects, with 10gb textures with the .max file being 160mb and taking 15minutes to open on a 4.6ghz 8350 ;]
I deleted everything, that was hidden during the render. Now it opens in almost instantly ;_;
No textures, obviously.

Settings were default, expect max ray intensity 4.0, light samples multiplier 4.0; the precalculation of UHD cache were done with 30 precomp. Amount (stopped at 200k anyways) and 4096 record quality. There were two caches, one for each room. Saved out and rendered the 150 frames with one and the rest with the other .hdc. Precomp was calculated by placing the camera in a corner and making it wideangle, so it covers all of the room.

https://corona-renderer.com/bugs/view.php?id=731

edit: when I switched to dailies there was a VERY strange bug. The first time I saved after the switch, two Spotlights kept reseting their fallof values, each time I reloaded the scene O.o Brain-Computer fart I suppose...