PDA

View Full Version : What causes this render hickup?



digitalimagery
01-08-2014, 09:10 AM
Anyone whose responded to my previous post about speeding up glass renders knows I'm working on a logo surrounded by various levels of square glass panels (210 frame animation).

I had to stop my render to restart my system due to a hang in another app. I noted the last frame rendered, opened Render globals and set the next frame to render in the sequence, and restarted my system. When I resumed rendering and checked the seq. this morning it appears there's a jump in the animation. This is unfortunate due to the fact that it's taking 1.5 hrs per frame and it looks like this hick up is going to require an re-render (unless there's a fix).

I know I can start about two seconds prior to the hit and do a dissolve to the second half but this shouldn't happen. Any ideas?

Thanks
DI

digitalimagery
01-08-2014, 03:32 PM
I may be answering my own question here. After analyzing the image (frame) where the problem is introduced, it appears (render still in progress) the refraction index returned to it's default of 1.52 when I re-opened the file after saving. Is this possible? Does LW do that? Pretty frustrating if so. Why does it (if this is the case) not retain surface settings when you save the file?

Anyway, I changed it back to 1.3 (where I started me render) and although tough to tell until I get the test frame out of the oven that this is what caused the change in the images I crunched last night. I'll update this string if I learn anything.

DI

MarcusM
01-08-2014, 03:36 PM
Before close Layout for sure you save not only scene but also "Save all objects"? It save surface settings.

spherical
01-08-2014, 03:40 PM
No, it doesn't automatically reset the IOR.

2 Questions:
Why are you not using the IOR of glass for... glass?
After all of the suggestions and testing by people here, why is a frame still taking 1.5 hours?

digitalimagery
01-09-2014, 07:02 AM
No, it doesn't automatically reset the IOR.

2 Questions:
Why are you not using the IOR of glass for... glass?
After all of the suggestions and testing by people here, why is a frame still taking 1.5 hours?

The IOR for glass was giving me too much distortion in the glass blocks so I tweaked it to not distort the logo behind it so much.

As for render times I suspect because I have 200 blocks of glass, two area lights and two point lights the amount of light bouncing, refracting and reflecting and passing through multiple glass blocks is choking it.

After experimenting with many suggestions (I was hoping for a higher res image of the suggested node approach posted by someone to experiment with). As it was in Williams tut, I was still getting long test times and decided rather than invest the time experimenting I could launch the render as is (which I think is very acceptable) and let it crunch on a remote machine while I move onto other work with deadlines. I have to answer to management and am limited in the amount of time I have. They don't mind a spare machine crunching numbers in the corner as long as I'm physically jamming on client work..

I fully intend on doing more tests with the dielectric node but the first few renders were giving me a lot of black. Someone suggested I still had the old duplicated flipped polys (which I didn't) and it was just time to move off this and onto other work.

I have noted the comments and will pick up on my home machine where I left off. I suspect the black areas were from lack of environment or background imagery.

Thank you for your time. I value the expertise and opinions here.
DI

spherical
01-09-2014, 06:20 PM
Ah, understood. A good alternative to time-expensive area lights are Dpont's DP Flood and DP Infinite. You can set the diameter, which controls shadow softness and do per-light sampling to be set differently from the Unified Sampling value for all lights. In the case of DP Infinite, it can also be set to Hemisphere.

JonW
01-09-2014, 07:43 PM
When I do glass balustrade, also for windows, all the glass on the far side & underneath I make non-reflective. I still have the solid block of glass but only the front faces are reflective. The back & other faces which will never be see by the camera I make transparent but turn off reflection, even have 100% transparent if it doesn't affect other things. Even have surfaces that are not really going to be noticed non reflective.

Otherwise reflections will bounce back & forth & add render time.

I am sure you have done all this, but quiet often it is the bleeding obvious we all forget!

Are all points Merged (Merge Points).
Check the dodgy polys that the corners are all Welded.
"W" check statistics that there are no 1, 2 or Non-Planar Polys.
Triple Non Planar polys.

digitalimagery
01-10-2014, 07:02 AM
Ah, understood. A good alternative to time-expensive area lights are Dpont's DP Flood and DP Infinite. You can set the diameter, which controls shadow softness and do per-light sampling to be set differently from the Unified Sampling value for all lights. In the case of DP Infinite, it can also be set to Hemisphere.

I'll experiment with that, thank you. I don't mean to sound so defeated but between, being a somewhat newbie (i'm getting it) I've now discovered as my logo comes to a rest and the panels continue to move, I'm getting a flicker of poly's in the seamless background which I believe is the global illumination setting being too low (see attached). This means additional tests and a re render - Ahhhhhhh.

On the bright side. Could be a good opportunity to try the DP lights. Thanks for the tip!

A very grateful,
DI

digitalimagery
01-10-2014, 07:25 AM
When I do glass balustrade, also for windows, all the glass on the far side & underneath I make non-reflective. I still have the solid block of glass but only the front faces are reflective. The back & other faces which will never be see by the camera I make transparent but turn off reflection, even have 100% transparent if it doesn't affect other things. Even have surfaces that are not really going to be noticed non reflective.

Otherwise reflections will bounce back & forth & add render time.

I am sure you have done all this, but quiet often it is the bleeding obvious we all forget!

Are all points Merged (Merge Points).
Check the dodgy polys that the corners are all Welded.
"W" check statistics that there are no 1, 2 or Non-Planar Polys.
Triple Non Planar polys.[/QUOTE]

Thanks JonW,

Actually, if you look at the image in the previous post, I need those back sides to reflect in the logo and the other panels behind the logo. I suspect if it made a huge difference (and it may) this would be possible on the very top most, bottom most and back most outer planes of each panel (which I assume is what you mean).

A couple of questions:
1. Can I turn off the reflectivity on an individual panel if points are merged? If so how is that done (again I'm relatively new).

2. I don't believe I've merged points (intentionally) because I want the rounded edges on the panels to have a smooth setting and the flat faces to be "not smoothed." Can those settings be determined prior to merging points and be retained once merged?

3. Can you look at the attached and see if this looks right (might have to re render for other reasons anyway - may as well fix this and learn something in the process). I do have three point polys in the corners because they're rounded. The two point poly thing has always been a point of confusion in my learning curve (self taught) and I usually merge points and if there are any two point polys afterwards I just delete them, see if it affected my model and if not move on. Am I missing anything there?

Thanks to everyone for holding my hand on this tricky self induced torturous LW project. Maybe in the future a time sensitive project involving glass will be put on my plate and this information will be what saves the day.

Lastly, hate to be whining so much about render times but this is a 7 sec animation at 1080p so with 210 frames render time is an issue. Add to that the fact that on the second half I intend to re render it with no forward panels so I can fade them off in AE means additional rendering. I'm aware I can fade them off in LW but this gives me more options on fade times and the time I start the fade. Yes I'm insane (if you're wondering).


Most grateful
DI


119198[QUOTE=JonW;1362172]

DI

spherical
01-10-2014, 08:27 PM
A couple of questions:
1. Can I turn off the reflectivity on an individual panel if points are merged? If so how is that done (again I'm relatively new).

2. I don't believe I've merged points (intentionally) because I want the rounded edges on the panels to have a smooth setting and the flat faces to be "not smoothed." Can those settings be determined prior to merging points and be retained once merged?

Merging points doesn't make any difference to what can and cannot be a) a different material, b) smoothed or not. Merge the points (sometimes automatic doesn't catch them all, so set a .5 mil fixed distance to check if any more are eliminated). Select the polys themselves that you want to do something different with and assign a different material name to them, then change the parameters.


3. Can you look at the attached and see if this looks right (might have to re render for other reasons anyway - may as well fix this and learn something in the process). I do have three point polys in the corners because they're rounded. The two point poly thing has always been a point of confusion in my learning curve (self taught) and I usually merge points and if there are any two point polys afterwards I just delete them, see if it affected my model and if not move on. Am I missing anything there?

No, that's what I do. Also check for points on zero, one or two polys and edges on only one poly or on more than two polys. Sometimes points on one or two and edges on one are valid, so don't just nuke them without checking.

What I'd really like to look at is the model of one set of those glass blocks to see what's going on there.

digitalimagery
01-13-2014, 07:28 AM
Merging points doesn't make any difference to what can and cannot be a) a different material, b) smoothed or not. Merge the points (sometimes automatic doesn't catch them all, so set a .5 mil fixed distance to check if any more are eliminated). Select the polys themselves that you want to do something different with and assign a different material name to them, then change the parameters.



No, that's what I do. Also check for points on zero, one or two polys and edges on only one poly or on more than two polys. Sometimes points on one or two and edges on one are valid, so don't just nuke them without checking.

What I'd really like to look at is the model of one set of those glass blocks to see what's going on there.

Heck you can see the whole darned thing. I'd be interested if anything else here is causing this flickering. Let me know if you need to see the animation and/or files so I can determine what's causing the slow render times as well.

Here's the model. I know the points at the tip of the S were a problem. Someone else graciously fixed the overlaps I got when I rounded the text.

Thanks so much. BTW I'm on LW 10.1
DI

digitalimagery
01-13-2014, 02:39 PM
Ah, understood. A good alternative to time-expensive area lights are Dpont's DP Flood and DP Infinite. You can set the diameter, which controls shadow softness and do per-light sampling to be set differently from the Unified Sampling value for all lights. In the case of DP Infinite, it can also be set to Hemisphere.

Spherical,

Thanks for the tip on the lights. Really liking the render times I'm getting (on a separate project) with these. Much faster.

DI

spherical
01-13-2014, 02:54 PM
Heck you can see the whole darned thing. I'd be interested if anything else here is causing this flickering. Let me know if you need to see the animation and/or files so I can determine what's causing the slow render times as well.

Yes, the scene file would be good, so we can see how your cameras, lights and GI are set up. I'll look at the models. Thanks.

digitalimagery
01-13-2014, 03:05 PM
Yes, the scene file would be good, so we can see how your cameras, lights and GI are set up. I'll look at the models. Thanks.

Please let me know If I left anything out. My day ends soon so I may not be able to respond immediately. Thanks.
DI


119275

spherical
01-13-2014, 03:30 PM
Thanks. I'm downloading 10.1 to re-install it (really cool that LW3DG keeps these earlier versions accessible), so as to be under the same constraints and not inadvertently introduce something that a later version may be capable of if I send a file back. is your 10.1 32-bit or 64-bit? Any plugins you rely upon?

MarcusM
01-13-2014, 03:31 PM
If you want have nice looking rounded edges in glass boxes you can do them during creating Box. In numeric window are options for that. Other solution is use round tool what you already done.
Turn on numeric window "n", make box tur on rounder tool, click "n" and you will see settings, check Mesh density: high.
Remember that on LW 10 in this rounding type you will have overlapping polygons on corners, this is bug, fixed in new LW.

In logo object ypu have 7 - 1 Vertex polygons.

If you have problem load objects from last LW: https://www.lightwave3d.com/assets/plugins/entry/input-node/

digitalimagery
01-13-2014, 03:47 PM
Thanks. I'm downloading 10.1 to re-install it (really cool that LW3DG keeps these earlier versions accessible), so as to be under the same constraints and not inadvertently introduce something that a later version may be capable of if I send a file back. is your 10.1 32-bit or 64-bit? Any plugins you rely upon?

It's 64 but I didn't expect you to go through all that. It's appreciated but not required. Really.

DI

digitalimagery
01-13-2014, 03:50 PM
If you want have nice looking rounded edges in glass boxes you can do them during creating Box. In numeric window are options for that. Other solution is use round tool what you already done.
Turn on numeric window "n", make box tur on rounder tool, click "n" and you will see settings, check Mesh density: high.
Remember that on LW 10 in this rounding type you will have overlapping polygons on corners, this is bug, fixed in new LW.

In logo object ypu have 7 - 1 Vertex polygons.

If you have problem load objects from last LW: https://www.lightwave3d.com/assets/plugins/entry/input-node/

Thanks MarcusM, This is turning out to be a little more trouble than I realized but I'm learning a lot about things I've avoided for these reasons. This community is such an amazing resource.

I'm very grateful to everyone.
DI

spherical
01-13-2014, 04:58 PM
It's 64 but I didn't expect you to go through all that. It's appreciated but not required. Really.

Well, it is if I'm going to offer any relevant suggestions.

Why is the model 1120' wide?
There is a missing image. Seems to be an Image World. It was looking for it when I loaded the scene but can't recall the name right now. With it gone, I was getting dark refractions. Took a while to find that this was the cause. Simulated with Gradient Background and things returned to as expected.

Initial tests are returning good results. Turned off a few unnecessary settings and the times dropped:
GI: 1m 18s > 0m 35.3s
Render: 8m 34s > 7m 54s

This is using your original "old style" glass textures; before moving into Dielectric.

For whatever reason, on the 64-bit box the 10.1 Hub won't stay running. I'll have to manually transfer object surface settings in order to save them. 10.1 on the 32-bit box works as expected.

spherical
01-14-2014, 03:40 AM
OK. Here's the latest test. Too much to explain in full right now but in short:


Fixed the logo "H" front and rear polys.
Triangulated non-planars.
Made the glass block front and rear faces non-smoothing because they were not refracting correctly.
Glass is Dielectric with db+w Material Booster. IOR 1.8 to pump up the refraction.
Glass color is lighter and less saturated to allow absorption to work properly.
Chrome is Conductor.
Ray Limit is up to 12.
Unnecessary settings in GI are off.
GI Multiplier 50%.

GI Time = 11.3 seconds
Render time = 6 minutes 8 seconds
119282

Does it look exactly the same? No. It wouldn't. Now the refraction and reflection are correct. The glass has more life. From here, adjustments away from actual can be made to "displace" the logo letter parts seen through the glass, if that is desired. I don't know what the original aim was.

digitalimagery
01-14-2014, 07:03 AM
OK. Here's the latest test. Too much to explain in full right now but in short:


Fixed the logo "H" front and rear polys.
Triangulated non-planars.
Made the glass block front and rear faces non-smoothing because they were not refracting correctly.
Glass is Dielectric with db+w Material Booster. IOR 1.8 to pump up the refraction.
Glass color is lighter and less saturated to allow absorption to work properly.
Chrome is Conductor.
Ray Limit is up to 12.
Unnecessary settings in GI are off.
GI Multiplier 50%.

GI Time = 11.3 seconds
Render time = 6 minutes 8 seconds
119282

Does it look exactly the same? No. It wouldn't. Now the refraction and reflection are correct. The glass has more life. From here, adjustments away from actual can be made to "displace" the logo letter parts seen through the glass, if that is desired. I don't know what the original aim was.

HolyCrap that's insane! 6.8 minute render time? MG! That looks great. Doesn't need to be a match just trying to get a glass panels refracting the logo as they pass through look (and you nailed it). That render time is insane. The only concern I would have is those flickering hot spots that appear on "my" animation (which I can post if you'd like). I think it came from the default GI setting but you'd probably know more than I.

Not sure what that missing file is but I've been known to experiment with various backgrounds and spherical environmental maps and not delete them if I don't use them (because I don't know how to remove them from my list(s).

So where do we go from here?

Thanks soooooo much.
DI

digitalimagery
01-14-2014, 07:50 AM
OK. Here's the latest test. Too much to explain in full right now but in short:


Fixed the logo "H" front and rear polys.
Triangulated non-planars.
Made the glass block front and rear faces non-smoothing because they were not refracting correctly.
Glass is Dielectric with db+w Material Booster. IOR 1.8 to pump up the refraction.
Glass color is lighter and less saturated to allow absorption to work properly.
Chrome is Conductor.
Ray Limit is up to 12.
Unnecessary settings in GI are off.
GI Multiplier 50%.

GI Time = 11.3 seconds
Render time = 6 minutes 8 seconds
119282

Does it look exactly the same? No. It wouldn't. Now the refraction and reflection are correct. The glass has more life. From here, adjustments away from actual can be made to "displace" the logo letter parts seen through the glass, if that is desired. I don't know what the original aim was.

After re reading your message about the width, I suspect you didn't get the seamless background. It's probably way bigger than it needs to be and I'm curious if that's adding to the huge render times (and what the answer is to getting a seamless look that doesn't affect render times).

I've attached it in the event it was missing. I believe it was in the objects directory called "new_seamless.lwo" but I'm always revising those to try for a smoother transition between the vertical and horizontal planes and this attachment is my latest version.

Thanks again.
DI

119284

spherical
01-14-2014, 04:59 PM
The width I'm referring to is the glass blocks. The seamless cyc is 7,250 feet long. Each of the glass blocks in the panel is 65 feet, 6 inches square. This seems to be way huge. I usually model in real-world dimensions. Many times, scale doesn't make much difference but in the case of glass and using Dielectric, the absorption factor is impacted by the distance through the material that the rays must travel. As a result, not only did I run the Value of the glass color up to 255 and dropped the Saturation down to 56, but set absorption down to 0.1 to obtain a faint green hue.

new_seamless.lwo is in the objects folder but is not in the scene. There is a seamless layer in the logo lwo, and it shows up in the scene, but it doesn't have an image mapped on it. When I disabled Image World, the scene quit asking for the image. Just opened the original scene in a text editor and it's looking for:

Macintosh HD 2:Production Elements/e-maps/StudyEnviro_Lrg.jpg

If this is indeed intended to be mapped with Image World, it would be interesting to see if it impacts the render. When the image was missing, but IW still enabled, I was getting dark refractions even after enabling Gradient Background to simulate your environment so I could continue. Unchecking it allowed the background colors to show up in the glass. Never knew that having it enabled and not mapped with anything would block the overall background. Good thing to keep in mind.

To remove an image, first delete the callout in the Surface Editor for that surface and then open Image Editor, right-click the image name and Delete.

Haven't done anything with the animation yet to see what moves and what doesn't. Unsure whether the Caustics calculation is necessary. I had it on and off, but made so many different changes to get a good result, that I can't recall if it made a significant difference in the render.

That "H" required making the front and back faces into three polys, two vertical and one horizontal across the bridge, to cure the web-foot problem. There were a number of non-planars that I Tripled and there are still some Edges that are only on one poly, so should be removed for a clean model. Most of them are where the two arcs are cut.

BTW, the render time is not 6.8 minutes, it's 6m, 8s. When I first applied Conductor to the sub_text surfaces, I got 6 minutes flat. I'm going to revisit my settings today to certify what is on and what is off, as compared to the original scene, so you know what has an impact. Will also look at the blotches. Yes, those are dealt with in the GI settings. While I'm looking at that, view RebelHill's video (http://www.youtube.com/watch?v=0YFZ2av-BLg) on it.

If you're going to be doing much with these types of jobs, invest in an upgrade to 11.6. Way better on a number of counts.

spherical
01-14-2014, 10:15 PM
OK. Interesting finding today. Caustics can actually help a render complete faster; probably scene-dependent. In working options to squeeze out as much as possible, I first eliminated Hypervoxels and Volumetric Radiosity, as they were doing nothing to the image. Then went to disable Affect Caustics on the Seamless Light and the Logo Light and turned off Caustics in the GI panel. The GI calc went from 11.4s > 13.2s but the render time went from 6m 8s > 10m 44s! Seriously? Even with incurring the 23 seconds to trace the caustics from those two lights, the render time was far less.

Then decided to see if I could obtain similar results to the GI render by other means, as I have done numerous times previous on other projects. Added a DP Dome light at 75%, 18 samples and get a render time of 7m 14s with no possibility of Interpolated GI blotches flickering in the animation. GI and Dome Light images together for comparison:
119296 119297

If you have the frame number where the blotches appeared in an earlier posted image, I can try to reproduce, then adjust GI settings to eliminate them for a comparison to the non-GI render time. I'm betting that upping the RPE enough to eliminate the blotches will push the time up past the Dome Light version.

OK, ran 600 RPE and got 7m 1s. 1000 RPE got way bloomed-out highlights in the glass reflections (for whatever reason) so not acceptable. 800 RPE came in at 7m 26s and looked exactly the same as 100 RPE and 600 RPE, but would presumably diminish to an acceptable level the blotches in GI. All things considered, I'd eliminate the possibility and go with the DP Dome light non-GI approach.

Here's frame 210:
119301

spherical
01-15-2014, 04:25 AM
Couldn't stand it anymore and went into the logo and cleaned up the geometry. Got rid of all the cruft (orphaned points, edges) and the webfoot in the sub_text "R". Model renders cleaner now. Cap "S" now exhibits some cool stuff, so running an animation render to see it all.

digitalimagery
01-15-2014, 11:11 AM
The width I'm referring to is the glass blocks. The seamless cyc is 7,250 feet long. Each of the glass blocks in the panel is 65 feet, 6 inches square. This seems to be way huge. I usually model in real-world dimensions. Many times, scale doesn't make much difference but in the case of glass and using Dielectric, the absorption factor is impacted by the distance through the material that the rays must travel. As a result, not only did I run the Value of the glass color up to 255 and dropped the Saturation down to 56, but set absorption down to 0.1 to obtain a faint green hue.

new_seamless.lwo is in the objects folder but is not in the scene. There is a seamless layer in the logo lwo, and it shows up in the scene, but it doesn't have an image mapped on it. When I disabled Image World, the scene quit asking for the image. Just opened the original scene in a text editor and it's looking for:

Macintosh HD 2:Production Elements/e-maps/StudyEnviro_Lrg.jpg

If this is indeed intended to be mapped with Image World, it would be interesting to see if it impacts the render. When the image was missing, but IW still enabled, I was getting dark refractions even after enabling Gradient Background to simulate your environment so I could continue. Unchecking it allowed the background colors to show up in the glass. Never knew that having it enabled and not mapped with anything would block the overall background. Good thing to keep in mind.

That image was used as an environmental map (spherical) for the arched tube like parts of the logo I believe). As for the seamless model that was the actual background and nothing was mapped to it. The grey looks close enough that it's splitting hairs but in the early frames (see early image postings) there was reflection showing in it. I attached that .jpg if you need/want it.


To remove an image, first delete the callout in the Surface Editor for that surface and then open Image Editor, right-click the image name and Delete.

Thanks. I needed that.


Haven't done anything with the animation yet to see what moves and what doesn't. Unsure whether the Caustics calculation is necessary. I had it on and off, but made so many different changes to get a good result, that I can't recall if it made a significant difference in the render.

That "H" required making the front and back faces into three polys, two vertical and one horizontal across the bridge, to cure the web-foot problem. There were a number of non-planars that I Tripled and there are still some Edges that are only on one poly, so should be removed for a clean model. Most of them are where the two arcs are cut.

Thanks, this is still a confusing part of modeling to me and It's hard finding tuts that cover if/then problem areas in models with these issues.


BTW, the render time is not 6.8 minutes, it's 6m, 8s. When I first applied Conductor to the sub_text surfaces, I got 6 minutes flat. I'm going to revisit my settings today to certify what is on and what is off, as compared to the original scene, so you know what has an impact. Will also look at the blotches. Yes, those are dealt with in the GI settings. While I'm looking at that, view RebelHill's video (http://www.youtube.com/watch?v=0YFZ2av-BLg) on it.

I understood the time you quoted, just wrote it wrong. I actually bought that tutorial (very well done and in depth). That's what made me aware of the whole GI issue. I thought my settings were high enough to avoid those (glass is a tricky beast) and was worried (based on comments in that tut) that I'd increase render times even further if I cranked them up further.


If you're going to be doing much with these types of jobs, invest in an upgrade to 11.6. Way better on a number of counts.

I do pretty much all logo work but this is a pet project for corporate. They make the call on upgrades. Not doing the actual work, they look at finished product (with no design experience whatsoever) and compare it to bottom line and don't see the value in the investment (bean counters). I've been requesting it regularly since it came out.

So while you're being so generous, what is a good base size to build a logo model? I've been using a grid size of 10 m, importing my eps logo and scaling it to fill the back view (working in HD formats). Another thing I question is, I downloaded a lot of the node surfaces from [URL="http://www.presetcentral.com/standardscene/"] and noticed I should be set to 10 cm not mm. I'll make adjustments to size based on that unless you can offer anything to the contrary.

Thanks so much for all you're doing on this project. I fully intend to "pay it forward" when I have something to offer on this forum.
DI

119307

digitalimagery
01-15-2014, 11:19 AM
Thank you. Still getting these finer details of modeling. With the help of this forum I've made some great strides.

DI

digitalimagery
01-15-2014, 11:31 AM
OK. Interesting finding today. Caustics can actually help a render complete faster; probably scene-dependent. In working options to squeeze out as much as possible, I first eliminated Hypervoxels and Volumetric Radiosity, as they were doing nothing to the image. Then went to disable Affect Caustics on the Seamless Light and the Logo Light and turned off Caustics in the GI panel. The GI calc went from 11.4s > 13.2s but the render time went from 6m 8s > 10m 44s! Seriously? Even with incurring the 23 seconds to trace the caustics from those two lights, the render time was far less.

Then decided to see if I could obtain similar results to the GI render by other means, as I have done numerous times previous on other projects. Added a DP Dome light at 75%, 18 samples and get a render time of 7m 14s with no possibility of Interpolated GI blotches flickering in the animation. GI and Dome Light images together for comparison:
119296 119297

If you have the frame number where the blotches appeared in an earlier posted image, I can try to reproduce, then adjust GI settings to eliminate them for a comparison to the non-GI render time. I'm betting that upping the RPE enough to eliminate the blotches will push the time up past the Dome Light version.

Frame 155 is the one I posted and circled.


OK, ran 600 RPE and got 7m 1s. 1000 RPE got way bloomed-out highlights in the glass reflections (for whatever reason) so not acceptable. 800 RPE came in at 7m 26s and looked exactly the same as 100 RPE and 600 RPE, but would presumably diminish to an acceptable level the blotches in GI. All things considered, I'd eliminate the possibility and go with the DP Dome light non-GI approach.

I'm so in the learning curve I didn't realize this was an option. So GI isn't affected if I use a dome light? Only affects area lights? What about those DP Lights we discussed earlier?


Here's frame 210:

119301

Thanks again,
DI

spherical
01-15-2014, 03:02 PM
I'm so in the learning curve I didn't realize this was an option. So GI isn't affected if I use a dome light? Only affects area lights? What about those DP Lights we discussed earlier?

No. You turn GI off.

digitalimagery
01-15-2014, 03:06 PM
No. You turn GI off.

Thanks - good to know. Learning learning.

DI

spherical
01-15-2014, 03:41 PM
Thanks, this is still a confusing part of modeling to me and It's hard finding tuts that cover if/then problem areas in models with these issues.

What you look for when you think the model is done are:


Points on zero polys
Points on only one poly
Points on only two polys
Edges on only one poly
Edges on more than two polys
Polys having only one point
Polys having only two points


SOME of these can be valid, depending upon the geometry in question.


A single poly, as an object, will have edges not attached to any other polys. IOW, a single square poly will have 4 edges not attached to any other polys. Put a hole in it and it will have more edges not attached to other polys, depending upon the number of sides that the hole has. The number will increase by the circle segments +1. This is the edge that locates the point on the hole and attaches it to the perimeter.
A volume shape with a hole through it will have two edges attached to only one poly each; the perimeter and hole connecting edges. All other edges will be on two polys.


We can use the logo as an example. The logo arcs, where they were cut, were where the points and edges had a problem. Check for points on one poly and you'll see what I mean. Lots of points not having intersections. The cutting shape had a different poly density from the arcs, so there are many more segments around the cylinder than there are polys that form the cut end. To get rid of this, I deleted the end polys, searched for and welded all points at intersections on the now open end of the arc and welded any points on the edges that were not at an intersection to the closest intersection point, then deleted the 2-point polys that were left. Now with only one point at each intersection and no points in between intersections, I created polys by point selection and assigned the correct surface to them. This got rid of a lot of the orphaned edges.

Orphaned edges are the most difficult to deal with. They are usually the result of bad modeling practice and/or an operation gone wrong, due to a unique circumstance. It happens to everyone. What I usually end up doing is deleting the polys that are on both sides of the problem edge(s) and then manually recreating the polys by selecting points. Sometimes just cutting and re-pasting the surrounding poly set, and then merging points will clear the problem but not always.

digitalimagery
01-15-2014, 03:51 PM
What you look for when you think the model is done are:


Points on zero polys
Points on only one poly
Points on only two polys
Edges on only one poly
Edges on more than two polys
Polys having only one point
Polys having only two points


SOME of these can be valid, depending upon the geometry in question.


A single poly, as an object, will have edges not attached to any other polys. IOW, a single square poly will have 4 edges not attached to any other polys. Put a hole in it and it will have more edges not attached to other polys, depending upon the number of sides that the hole has. The number will increase by the circle segments +1. This is the edge that locates the point on the hole and attaches it to the perimeter.
A volume shape with a hole through it will have two edges attached to only one poly each; the perimeter and hole connecting edges. All other edges will be on two polys.


We can use the logo as an example. The logo arcs, where they were cut, were where the points and edges had a problem. Check for points on one poly and you'll see what I mean. Lots of points not having intersections. The cutting shape had a different poly density from the arcs, so there are many more segments around the cylinder than there are polys that form the cut end. To get rid of this, I deleted the end polys, searched for and welded all points at intersections on the now open end of the arc and welded any points on the edges that were not at an intersection to the closest intersection point, then deleted the 2-point polys that were left. Now with only one point at each intersection and no points in between intersections, I created polys by point selection and assigned the correct surface to them. This got rid of a lot of the orphaned edges.

Orphaned edges are the most difficult to deal with. They are usually the result of bad modeling practice and/or an operation gone wrong, due to a unique circumstance. It happens to everyone. What I usually end up doing is deleting the polys that are on both sides of the problem edge(s) and then manually recreating the polys by selecting points. Sometimes just cutting and re-pasting the surrounding poly set, and then merging points will clear the problem but not always.

Thanks that clarify's a lot of issues I've had with my models (just didn't know what to do about it). I suspect they all have some sort of problem but they look fine rendered so I move on. I do want to know how to model properly though. One day this is going to bite me.

Thanks for the detailed explanation.
DI

spherical
01-15-2014, 05:35 PM
Yes, they may "look fine when rendered" but often when you get a model to be sound, it'll look great. Sometimes, it doesn't show in the render, other times it does. I believe the "S" in the logo will illustrate what I mean, once the frames are all completed. For whatever [email protected]$$ reason, my workstation decided that 100% load on 12 cores wasn't enough, so it went to sleep in the middle of the night with only 1/3 of the sequence completed. I expected it to be asleep when I got to the studio, but with the sequence all finished. Not.

To add insult to injury, it appears that the render timer kept running. At 50% rez, the frames are taking 2m 9s. Should've been done in 7.5h but at 9.5h it was 30%.

digitalimagery
01-16-2014, 06:45 AM
Yes, they may "look fine when rendered" but often when you get a model to be sound, it'll look great. Sometimes, it doesn't show in the render, other times it does. I believe the "S" in the logo will illustrate what I mean, once the frames are all completed. For whatever [email protected]$$ reason, my workstation decided that 100% load on 12 cores wasn't enough, so it went to sleep in the middle of the night with only 1/3 of the sequence completed. I expected it to be asleep when I got to the studio, but with the sequence all finished. Not.

To add insult to injury, it appears that the render timer kept running. At 50% rez, the frames are taking 2m 9s. Should've been done in 7.5h but at 9.5h it was 30%.

I usually attribute things like that being the result of the "crisis detector" which lives in every professional animators computer (regardless of make, model or OS). In short, everyone's system runs perfectly until there's a crisis - be it tight deadline, important client, high profile animation (whatever). THAT"S when the crisis detector knows you need something yesterday and BAM the SHTF.

Not the case here but all the signs of it. Fortunately there's no hard deadline on this one (at least nothing real soon) which is why I opted to attempt glass in the first place. Been wanting to experiment.

I appreciate all the time you've spent helping on this. The information you've shared is invaluable.

DI

spherical
01-17-2014, 05:33 AM
OK. Here's the animation test render, so I could see how this is set up.


Fixed logo object to be water-tight; removing extraneous cruft in the geometry, so that smoothing would work reliably.
Dielectric material for glass blocks.
Conductor for small lettering.
No GI.


http://glasssculpture.org/client/LightWave/sw-logo-test.html

digitalimagery
01-17-2014, 07:09 AM
OK. Here's the animation test render, so I could see how this is set up.


Fixed logo object to be water-tight; removing extraneous cruft in the geometry, so that smoothing would work reliably.
Dielectric material for glass blocks.
Conductor for small lettering.
No GI.


http://glasssculpture.org/client/LightWave/sw-logo-test.html

Thanks for all the tips on how to get render times down. Love to see a frame in high res. It's hard to read the real quality on a small compressed animation.

DI

spherical
01-17-2014, 07:03 PM
119374 119375 119385

Be sure to click on them two or three times, in order to get them to full rez.

Then, here's something that you can go to your boss with. I brought the scene into 11.6.1 and only made setting adjustments that are specific to Unified Sampling: Using the per-light samples capability of DP Dome, and setting the light and camera samples to pretty much generic values. Time to render frame 093 at full resolution (the third image) dropped to 2m 58s. This is the time I was getting when rendering at 50% size and lesser quality values. There are some slight variations in the specular highlights and refractions that no doubt can be brought out. This was a simple 5 minute reality check, as I was limiting myself to v10.1 previously.

Sometimes, there is a serious false economy in not investing in new technology. But, quite often, there is no logic involved when dealing with bean counters. They don't get the Big Picture, because they are myopically focused on the beans.

digitalimagery
01-20-2014, 07:13 AM
119374 119375 119385

Be sure to click on them two or three times, in order to get them to full rez.

Then, here's something that you can go to your boss with. I brought the scene into 11.6.1 and only made setting adjustments that are specific to Unified Sampling: Using the per-light samples capability of DP Dome, and setting the light and camera samples to pretty much generic values. Time to render frame 093 at full resolution (the third image) dropped to 2m 58s. This is the time I was getting when rendering at 50% size and lesser quality values. There are some slight variations in the specular highlights and refractions that no doubt can be brought out. This was a simple 5 minute reality check, as I was limiting myself to v10.1 previously.

Sometimes, there is a serious false economy in not investing in new technology. But, quite often, there is no logic involved when dealing with bean counters. They don't get the Big Picture, because they are myopically focused on the beans.

Thanks Again spherical. Yes it's hard to get mgmnt to bite on artistic logic (they think we just want new software) it's like pulling teeth. This coming year looks like it will be a good one for business and that "may" grease the wheels. They will say if glass takes so long to render then don't do glass (as opposed to forking over the dollars. One can only hope.

As for the render tests, I can easily live with the render times you got in 10.1. As impressive as the time is on the 11.x version the grain is unacceptable and cranking up the light quality to eliminate it would probably bring that time back into a similar range.
Very nice quality on those frames, thanks, looks real good.

DI

spherical
01-20-2014, 03:19 PM
What grain? Which image are you looking at? If you're talking about the animation, it was done at a lower AA, no AS, 50% size and then scaled down 33% in AE. I wasn't concerned about ultimate quality. I wanted to see how you set up the animation and what refractions did. Besides, motion blur, will not only eliminate any you may get but will also lessen the amount of AA/AS you need to apply.

digitalimagery
01-20-2014, 03:21 PM
What grain? Which image are you looking at? If you're talking about the animation, it was done at a lower AA, no AS, 50% size and then scaled down 33% in AE. I wasn't concerned about ultimate quality. I wanted to see how you set up the animation and what refractions did.

I was referring to the tiny speckles on the third image. If you click between the second and third you'll see what I mean. They're subtle but looks similar to a noise filter in PS.

DI

spherical
01-21-2014, 12:13 AM
Unacceptable... Well, most of us doing serious animation will add grain to make it more like film and less pristine. It's similar to the difference between vinyl source and tube amplification as opposed to digital everything. The audiophiles here will get the reference.

Point is, motion blur would more than handle the left over artifacting not smoothed by AA/AS. However, I decided to depart from the minimal 10.1 <> 11.6 conversion, which is all that I was attempting to show (apples and apples, as much as possible), and see just what could be done in 11.6. I could probably go farther into it, but this pretty much tells the story:

1st image, 2m 47s - 11s down from the 2m 58s previously posted.
2nd image, 2m 0s. I was going to go with this when I first saw it but figured that it would be unacceptable. Again, at 30 frames per second. MB would more than handle the minimal noise in the logo highlight.

119463 119464

That which made the biggest difference is getting rid of the expensive 2-sided area lights; replacing them with arrays of DP Floods, then adjusting down the global light samples, per-light DP Flood samples, Min/Max camera samples, and AS threshold. Seeing as how your original frames were taking 1 hour 39 minutes, I'd say that these improvements, noise in the previously posted image that was unacceptable notwithstanding, are a significant improvement.

digitalimagery
01-21-2014, 06:45 AM
Unacceptable... Well, most of us doing serious animation will add grain to make it more like film and less pristine. It's similar to the difference between vinyl source and tube amplification as opposed to digital everything. The audiophiles here will get the reference.

Point is, motion blur would more than handle the left over artifacting not smoothed by AA/AS. However, I decided to depart from the minimal 10.1 <> 11.6 conversion, which is all that I was attempting to show (apples and apples, as much as possible), and see just what could be done in 11.6. I could probably go farther into it, but this pretty much tells the story:

1st image, 2m 47s - 11s down from the 2m 58s previously posted.
2nd image, 2m 0s. I was going to go with this when I first saw it but figured that it would be unacceptable. Again, at 30 frames per second. MB would more than handle the minimal noise in the logo highlight.

119463 119464

That which made the biggest difference is getting rid of the expensive 2-sided area lights; replacing them with arrays of DP Floods, then adjusting down the global light samples, per-light DP Flood samples, Min/Max camera samples, and AS threshold. Seeing as how your original frames were taking 1 hour 39 minutes, I'd say that these improvements, noise in the previously posted image that was unacceptable notwithstanding, are a significant improvement.

Thank you for the production lesson. You clearly have years of experience and I'm lucky to have someone like you spend your valuable time enlightening me on the finer aspects of video production. Your knowledge of LW is truly impressive.

DI

spherical
01-21-2014, 08:05 PM
No it isn't. I'm learning, just like you. Helping you, if I did, taught me a lot and I am happy to have had the opportunity. When I went in to research the noise issue, eventually I was delighted to have developed a method that would give me completely smooth areas and a lower time. Went through more than a few trial and error sessions to get it. I figured that getting both would not be possible, but I kept at it anyway, just to see what could be achieved.

Just to be clear, I didn't particularly appreciate the tone of your last couple of posts. They differed from those previous. If that wasn't your intent, they could have been put in other terms, so that there would be no misunderstanding of perception. Still, I do hope that you can gather the information that we covered and make your animations in a more efficient way. If you can take this last bit of work and use these examples to now show your superiors where you can save major amounts of time (doesn't only apply to glass) at the same quality by granting an upgrade to 11.6, so much the better. I'd even suggest that if they still give you the stiff arm, get an upgrade yourself, with the stipulation that you take it with you when you leave. A lot has changed in LW since 10.

digitalimagery
01-23-2014, 08:40 AM
No it isn't. I'm learning, just like you. Helping you, if I did, taught me a lot and I am happy to have had the opportunity. When I went in to research the noise issue, eventually I was delighted to have developed a method that would give me completely smooth areas and a lower time. Went through more than a few trial and error sessions to get it. I figured that getting both would not be possible, but I kept at it anyway, just to see what could be achieved.

Just to be clear, I didn't particularly appreciate the tone of your last couple of posts. They differed from those previous. If that wasn't your intent, they could have been put in other terms, so that there would be no misunderstanding of perception. Still, I do hope that you can gather the information that we covered and make your animations in a more efficient way. If you can take this last bit of work and use these examples to now show your superiors where you can save major amounts of time (doesn't only apply to glass) at the same quality by granting an upgrade to 11.6, so much the better. I'd even suggest that if they still give you the stiff arm, get an upgrade yourself, with the stipulation that you take it with you when you leave. A lot has changed in LW since 10.

I come to this forum to get assistance learning a complex program for which I've never been formerly trained. But if you want to talk about "tone" it's rather insulting to be talked down to by someone you turn to for help. Your words were "Unacceptable... Well, most of us doing serious animation will add grain to make it more like film and less pristine."

If you find my personal opinion of grain being an insult I'm sorry, I'm entitled to my own opinion on how "my" animations look and what I'm after. If grain was such an important part of what you look for I wonder why it isn't on all the previous images you posted (or the ones on your own web site).

If your comment about "most of us doing serious animation" is not talking down to someone I don't know what is. I don't live in California or work on major blockbuster films (do you?) because I made the decision early on in my "professional production career" to have a life and be part of my children's lives (and I wouldn't trade that for burning out or my name on a credit roll for a million dollars). I have however been in the production business for over 35 years (professionally employed) and have won numerous national awards for my work (more than 10). I could go on but this is not the place to be slamming each other.

I appreciated all the work you put into helping me understand a program I'm new too but your assumptions of my abilities, my experience or my opinion of what "I" want in my art were neither called for or appreciated. You should be more careful of what you assume. There are some really really talented people in this forum. Most are respectful and more importantly, humble.

Blessings,
DI

djwaterman
01-23-2014, 10:58 AM
What the?

spherical
01-23-2014, 03:40 PM
Really...


There are some really really talented people in this forum. Most are respectful and more importantly, humble.

All it was, to me, was a flavor of looking a gift horse in the mouth. I put all those hours in helping you (and gaining experience myself, so all is not lost) and get "unacceptable" as a response. A little bit of noise in flat areas, as I said, can be handily accommodated by Motion Blur, where the AA/AS settings can usually be relaxed. A still image is different and the AA/AS values need to be way higher. Simple fact. You're evidently an accomplished animator, so you should be able to gather the things that I shared with you and do even better than I. Quite frankly, I didn't follow my own instructions and at 50% size in the Image Viewer there was no grain evident. At 100%, as you pointed out, it was, so I went back in and found a way to completely eliminate it and drop the render time to boot.

As for the statement about adding film grain, it was a statement, nothing more. It is done all the time and was a relative reference to the alternative of obsessing about grain in stills, when in the animation it won't make much difference, if any, and sometimes is purposefully inserted. It was intended to put perspective on the situation. If you want to make it into something more than it is, that is your call.

Insults, name-calling and comments about one's character get zero around here. Lighten up.