Indirect Light and scene scale.
tbwoq
Posts: 238
A while ago, the Carrara native content was reworked to better match the scale of objects in scene magnitudes. I noticed some of the GI scenes now take much MUCH longer to render on my system. Some popular scenes like the indoor House Day(etc.) used to render faster. This scene had Sky light and Indirect Light checked. I remembered some of the scenes used to be a small scale magnitude. Sure enough, when I scaled it down, it rendered faster.
For instance. On my system, the House Day scene(default) takes approximately 2hrs:25min. If I group and scale the scene down to 1% of the overall scale, then reset the group to origin(0,0,0), it now takes 1min:49seconds. While the quality degrades slightly(which appears mostly related to Interpolation being checked), its no where near that much of a difference to waiting 2 hours(example images below). At first it appeared to be the scale differences maybe linked to photon mapping etc. However, I later find this is not the case.
Sometimes when rendering a scene, I noticed depending on how I moved the objects, the render times were very inconsistant. By accident, I found that if you move a single object out to great distances, the indirect light changes in the similar way as scaling down a scene does. So the indirect light rendered much faster with this method as well, only now the quality can be inconsistant.
Example. Load the 'House Day' scene from the browser. Insert a sphere anywhere in the scene. Move or translate the sphere to 100k(ft) distance or more on the Z axis. Render the scene. It renders darker, but still much faster. To further test, I experimented with other outdoor scenes. It works just as well and I get very realistic results with some fine tuning.
So, scaling an overall scene down or moving an single object to great distances affect the indirect lighting renders in a similar way. I might be missing something obvious or could this just be my setup. Anyone else experience similar results?
The first image takes approx. 1minute 49 seconds to render using the smaller scale. The second takes approx. 2 hour 25 minutes (EDIT:using the default medium scene). Both have the same default render settings.
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Hi there.
Whilst I haven't spent much time with GI (too slow for an animator), I would like to comment on scene size and render speeds.
Scene size affects lighting. The smaller you go (I've gone to 1%, but often I'm in 10%), the more likely shadows might become a problem (i.e. they will not be there). Usually it's ok.
I think there are many bugs in lighting. You can almost think of this as a "feature". For example the Glow on a bulb just isn't big enough to do a decent street light at night (at full size), take it down to 10% and voila! Just right.
I've also had problems with Distant lights at full size on very large scenes. They run out! So again, rescaling the models to 10% and now the distant light goes further.
However your discovery is excellent!
Perhaps the secret now then is to render a frame at full size with render settings set to something flimsy (like no anti-ailising, low object and shadow accuracy) to get the rough idea of what it *should* look like ( I say this with a pinch of salt as the only real guide are reference photos imho), then reduce the size to knock the frames out.
GI plays up with animation though. I think this particular scene causes problems from memory (the additional lights for the lamps etc vary in performance according to where the camera is - do a pan from left to right to see what I mean).
In summary (as I feel I'm going on a bit). I prefer to see what it should look like, then fake it all with fake indirect lighting (usually tubes, but bulbs are ok sometimes) PLUS scene size. That way you can get the lights to behave differently as explained with glow above.
Hi Sci Fi Funk.
Indirect lighting or maybe the render engine seems to have scene scaling limits. Still experimenting to find out what the limits are. I did notice the size of glow channel shaded objects will make more or less light but had not tried it at various scene scales yet.
When scaling indirect light scenes down, it does degrade the quality in some areas, but I can usually fine tune the scene or render settings to make it look better. At these massively decreased render times its worth it.
With indirect light animations, the light ends up being calculated different for each frame(afaik). This causes flickering shadows or light in some areas. If the camera is the only thing that moves, the Irradiance Map save feature should be able to avoid it. I haven't tested with only a Sky Light yet.
The main reason I wanted indirect lighting was for more realistic scenes. I didn't like the way some HDRI and Sky Light look as they don't bounce color like indirect lighting does. Having only a single core system(atm), I had to find ways to decrease render times. With the above scene scaling method, I can now test or even final render scenes without the massive render times.
Below image is Howies Snow Scene with indirect light and some atmosphere adjustments. One sun light, no sky light or background. Render time was around 1hr:23min. at (EDIT:default render settings from the scene except filter sharpness at 100% and IL at 145% intensity). Reduced the image for the forum, but should still give a good example.
I see. Thanks for your insights. I'll have a try with this at some point in the future.
We all want more realistic renders, and your kind revelation as to the time saving by scaling with GI is a great find.
I see. Thanks for your insights. I'll have a try with this at some point in the future.
We all want more realistic renders, and your kind revelation as to the time saving by scaling with GI is a great find.
Glad it helped, and thanks for your insights and tutorials as well. The irradiance map feature is also a huge time saver for some scene setups. I think its the same as the Global Illumination Multi-Pass element.
So.... Carrara is creating a scene-sized "grid" and filling it in with photon elements.... Then ignoring the areas without any geometry.... And the issue of quality is related to how many photon elements are creating the light map....
Scaling down, Carrara assumes fewer photon elements - a lower resolution "grid".
It makes sense. And you discovered how to control it: Send a primitive off into the distance to change the size of the scene....
This is amazing information!
So.... Carrara is creating a scene-sized "grid" and filling it in with photon elements.... Then ignoring the areas without any geometry.... And the issue of quality is related to how many photon elements are creating the light map....
Scaling down, Carrara assumes fewer photon elements - a lower resolution "grid".
It makes sense. And you discovered how to control it: Send a primitive off into the distance to change the size of the scene....
This is amazing information!Agreed - and I am very excited to try this for myself.
Steve, I think that, what you've been noticing with Distant lights (due to their infinite nature) that some elements of your scene were blocking them at full scale, but no longer when reducing the scene size. I could be wrong. Distant lights have an infinite path, so if you block that path anywhere along the grid (not just according to the placement of the light model, itself) it will create a shadow. If that shadow is large enough, it can (very easily) block the affects of the light altogether.
Back to the original topic, I'm really glad that Holly explained it in her way - as I was thinking along those same lines, when reading the discoveries in the original post, but was fumbling over an explanation! lol
That kind of time savings might just make animation and GI/IL a more useable endeavor and, like I said, I can barely wait to try it.
Once again,
thanks tbwoq! (and Holly and Steve)
This all seems to be known as it is mentioned in the C7 Manual on page 795.
Along with this there is a hint for saving memory.
For me the question now is:
Are these supposed to be features or just workarounds for bugs in the render engine?
Eddy
-holly wetcircuit.
No idea whats happening yet when moving the objects out miles from the origin. Content like Victoria 4.2 is especially bad. I tested 25 miles away and, wow, it does something strange to the mesh or render. There was a bug reported about this issue but I can't check the status. Don't know if it got fixed in 8.5. I'll continue researching and later, start testing photon maps.
-Dartanbeck.
There might well be some kind of limits or rendering precesion issues like holly said, but I'm now finding out it might have to do with all lighting. If the users with 8.5 say that the above mentioned bug with content is still happening, then there might be something in the render engine that possibly degrades the rendering precision on purpose. Why this would be, I'm not sure. The rendering precision is mentioned when creating a new document(scene).
-EddyMI.
It could well be a bug with how the render engine processes objects and scene scale or distances. Or it could just be a limitation of rendering precision in Carrara. I'm guessing at this point, that its something different from what is mentioned in the PDF manual.
Which manual? The one I have (revision G) talks about batch rendering on page 795.
I have the same Revision.
But note, that the page number in the document differs from that one that is shown in the reader because of the intro pages at the beginning of the manual. 2 unnumbered pages and 9 content pages with roman numbering and another unnumbered page. That makes the difference of 12.
Document page 795 + 12 = Reader page 807
Ugh me so dumb. Thanks.
I do not see this info on page 795...?
-holly wetcircuit.
No idea whats happening yet when moving the objects out miles from the origin. Content like Victoria 4.2 is especially bad. I tested 25 miles away and, wow, it does something strange to the mesh or render. There was a bug reported about this issue but I can’t check the status. Don’t know if it got fixed in 8.5. I’ll continue researching and later, start testing photon maps.
I am feeling especially dumb this morning... The mesh is deformed? Does anyone have a pic of this?
Here it looks like this...
Yes, that is the same page I have, but it is not any of the info listed above in tbwoq's post.
But you noticed it yourself in post #6:
So.... Carrara is creating a scene-sized "grid" and filling it in with photon elements.... Then ignoring the areas without any geometry.... And the issue of quality is related to how many photon elements are creating the light map....
Scaling down, Carrara assumes fewer photon elements - a lower resolution "grid".
It makes sense. And you discovered how to control it: Send a primitive off into the distance to change the size of the scene....
This is amazing information!
I don't see any mention of this in the manual.
As I understand it, the passage in the manual is talking about toggling off the visibility of a scene object, so its geometry is not added to the render calc. And suggesting that keyframing this property when this object is not calculated will drastically improve render speed.... This matches with my own observation that rendering in layers with alpha is drasticaly faster than rendering a subject and background in the same scene....
tbwoq (as I understand it, I have not tested) has discovered that under global illumination, shadow artifacts appear under certain scaling conditions - in otherwords, Carrara treats global illumination as a property with "scale". You can see the issue in his indoor pics above look around the recessed light fixtures in the ceiling, there are odd harsh shadows coming from an incorrect angle.... Also look between the sofa and the coffee table on the floor, the "high quality" render is smooth, but the "low quality" render looks as if there are several small lightbulbs not quite in alignment casting the shadow on the floor.
I don't really understand how Carrara figures out its GI, but it seems to be making a "photon element" <-- my bs term, that is a bunch of blobs of light floating in space, each more or less the average of the light that should be there.... If the blobs are very fine it all looks good but takes too long to render..... If the blobs are too big they show weird ashy patches of light and shadow but it takes much less time to render. Controling Carrara's GI has always been a bit of a mystery. Some people love it and some hate it - probably related to what kinds of scenes they typically make.... tbwoq's observation suggests that you can control the GI scaling by placing reference objects to change the size of the scene to deliberately change the scale of the blobs.</p>
I think. LOL
Okay, perhaps I read this wrong - as I was seeing this as a good thing.
I already know that figures, like V4 for example, with detailed texture maps render much faster at a distance than up close. You're not actually decreasing polygon count at all, unless you're using LOD, I guess - I still haven't figured out where to find LOD stuff....
What I was seeing from this is that, something like what Holly was saying, the decreased scale of the scene doesn't decrease the size of the grid - so that the grid (if that is truly what it is) becomes faster, since it isn't working as hard - due to the render requiring less pixels per item - even though the polygon count and map sized remain the same.
I'm performing a test right now in 8.5
We'll see how it turns out.
The test I wanted to perform didn't turn out. I took an existing scene and grouped it. Then uniformly scaled it down to 1% of the original size. Not everything scaled properly - so the test didn't work. I am now convinced that I 'was' reading this wrong! lol
The questions I have about this issue are:
Photon Map - is it similar to baking? If so then I am assuming that the photon map is basically all the geometry in the scene laid out flat sort of like a big UV map....
I would assume that complex geometry uses up more of the map than simple geometry..., but it could also just be a simple calculation of surface area - in that way the *size* and *scale* of the geometry is more important than the number of facets and polygons... (in otherwords, NOT like a UV map). Like the way a brain has more surface area than a similar-sized sphere, objects with lots of contortions have more surface area than smooth objects, regardless of the number of polygons used....
My point is: when you use these "spoiler objects" to change the size of the scene, it might make a considerable difference whether they are simple cubes or complex shapes. A simple shape may not use much of the photon map, while a convoluted shape may use more....
If you have Pro, isn't there a way to render the photon map with the Multi-Pass (as seen from the camera's POV)? That might be a much easier way to visualize what the photon "blobs" are doing, their scale and number....
-Dartanbeck.
Scaling some parented objects down may require you promote the grouping to master first or the objects will scale different. I had to do this for Howies snow scene or trees would completely disappear. The LOD and render smoothing features shouldn't be affecting this scaling method, but there is something similar going on.
-holly wetcircuit.
I'm not sure how the photons get setup in this scaling method yet, but haven't seen any set maps or limits(other than render setting values). After testing with another GI scene it appears when moving objects there are alot of other issues linked to scaling. I tried spline objects, massive distances, small and super sized objects, and even reversing normals but none affected the GI scene I was testing. Still researching.
Objects out of view do affect Howies snow scene though, and one of the mountains actually has part of it disappear in preview. So something is definately going on with the objects at distances. It could be something linked to the universes origin and/or limits on regular lights, which in turn affects the indirect lighting.
The Show Photon Map feature as I understand it is for adjusting indirect lighting details. An interior room might look good at certain settings, but a small object sitting on a table in the same room might not. Seeing the photon map overlay you can adjust the photons for details like that. The Multi-Pass Global Illumination works similar to irradiance map save feature. If you save an irradiance map and load it back, then render without lights, only the bounced lighting is visible in scene.
Promote the grouping to master. What do you mean by this?
Some objects and groups with parent child links have different setups. These sometimes won't scale down uniformly without undoing the link, ungrouping, changing the objects or propagate scale settings(etc.). Promote to Master creates a single master group(one instance) out of those groups or objects. Then use the Overall scale on that master group instance.
Its simlar to master objects. If you make a change to the master group, any duplicate of the original master will show the changes as well. You can still Jump In/Out of master groups to change and even animate objects. There are limitations though. Some modifiers won't work within a master group, replicated master groups won't render, and rigged objects can't convert to a master group.
This is a really interesting thread! I can verify the dramatic speed increase with the House Day scene, roughly speaking scaling to 10% gives a 10x speed increase and scaling to 1% gives 100x speed increase, but with deteriorating quality. I think this is broadly similar to decreasing the photon count in a scene (except that you cannot reduce this below 1000, so scaling gives effective access to lower rates).
I also tried this with a figure in Carrara Portrait Studio - the speed improvements were there but a lot less - a scene that normally rendered in 20 mins, reduced to 10% scale rendered in 15 mins and at 1% scale rendered in 10mins. There was a noticeable decrease in quality and at 1% the eyes looked black, like you get when you don't check the Light Through Transparency option in the GI panel.
So for figures this may be of restricted use, but for interiors, architectural etc it could provide a real boost in potential rendering speeds!
Hi PhilW.
I agree, there is something in the render engine that causes Carrara to adjust photons(?) settings at these scales. I'm still trying to figure it out so that I can find a balance between render time and quality.
I tested scaling with a Victoria 4.2 figure in scene. The eyes rendered up close OK but with the lower quality, but still not bad for the render time. I don't have Portrait Studio, but if you have indirect light checked, there should be much more of a speed increase than you had.
Architectural renders was one reason I wanted faster indirect light. Combined with irradiance maps, even my system might render some complex scenes now. At the least, some very nice test renders for an approximate lighting solution.
Carrara's render engine seems to be adjusting the way photons, and possibly lights in general, are rendered based on relative distances to origin, camera frame, and scene scales combined.
Here are a few things found so far.
>Photon map accuracy changes. This may account for most of the render speed increases, but its not the only reason.
>Object distances from the main camera frame affect lighting, even if their out of frame.
>Rendering artifacts(some extreme) occur when rendering far from the origin, especially for rigged objects.
>Lighting is generaly darker. Possibly due to photons counts being reduced and/or distance from origin.
>Lights that are scaled down with the scene might stop rendering at certain scales(this might have been what Sci Fi Funk was observing with distant lights).
I tested one other GI scene, and moving a separate object miles from the origin doesn't seem to change the lighting or photons. However, the spot light in that scene stopped working at certain scales. None of its settings had changed. Interesting that if you replace it with another spot light at that point, the new spot light works fine. So scaled lights appear to have limits. Sky Light alone seems unaffected in renders from scaling down, but atmospherics like fog, haze, volumetric clouds may not render well or at all.
The rendering artifact bug for objects at far distances from the origin, appears mainly to be connected to rigged objects or video cards. Even a single sphere with one rigged bone shows mesh type artifacts. After rendering, bringing the object(s) back to origin, the render problems stop, but not for some of the video display preveiws(like Flat or Wire). Not sure how all this affects the scaling method, but continuing to research.
Retested the House Day scene. I wanted to test blurry reflections with the scaling. Carpet objects(or floor) at 100% blurry reflection, Best, and raised the channel ray depth from 2 to 32. This usually sends my CPU into a dimension where time stands still, but it rendered in 40min. 31seconds.
That's a cool looking render - the blurred reflections add a lot, and anything that speeds the rendering of those is to be welcomed!
Very interesting stuff, my friend.
Sci Fi Funk's issue was the opposite. He was not getting what he wanted from his distant light until he reduced the scene scale - which leads me to believe that a building or some such may have been blocking its infinite pathway.
Hey, thanks for taking the time out to report all of your findings on this. I am so piqued on this subject it's making me wish I had more time to join you in these tests.
Right now I'm busy putting my modeling skills to the test in Carrara - and I'm not finding myself wishing for anything different than what I have to work with. It's a very complete set of tools and I love how they're organized about the well designed interface. Man, I love Carrara Pro! And having 8.5 along with DAZ Studio 4.6 Pro - the latest build - I'm having all manner of fun creating things for Genesis 2.
Maybe later I can take a break and experiment some in these regards. I certain hope so.
The "Photon Map accuracy" in Carrara change's the photon search radius (and maybe the search count?). The higher the percentage, the smaller the search radius. That does then mean more photons would be needed for good coverage.
The "Lighting quality" and "Accuracy" settings (below the photon settings), I am not completely sure, but those may control the final gather .