Light Leaks using extruded splines

[SandmanNinja]

3ds 2009 64-bit, Windows XP64, Daylight System

 

Hi...

 

I'm taking this class and the instructor has us modelling with extruded splines - which is great.

 

However, when I try to use photons, I get very obvious light leaks between the splines.

 

Here's my work flow:

01) put the daylight system in the scene

02) put the camera in a room with big windows

03) turn FG off and GI on, 2 photons, sampling radius of 10mm

This shows me where the photons are hitting

04) I turn the sampling radius up to 100mm

The photons now have what I call a 'Flower Pattern'

05) If the Flower Patterns do not overlap 'enough', I turn the sampling radius up to 250mm

06) I then increase the number of photons from 2 to 100

07) At this point, I usually have a basic lit scene

08) I turn on FG on low settings to even the splotchiness out

09) adjust the FG parameters until it looks good

 

However, I have used Editable Polys up til now and never got a light leak.

If I did, I just turned down the sampling radius from 500mm to 250mm or 100mm until it stopped sampling from the 'back side' of the wall where the light leak was coming from.

 

The splines were drawn out and extruded. On paper, measurement-wise, there are no gaps (at least none that I can see).

 

But they are separate objects - the upper walls, the courtyard walls, the floor, etc. Should they all be attached and then verts welded?

 

exterior

 

 

 

interior

[mattclinch]

that doesn't look like light leaking to me, more like you are smoothing over too few photon samples. i would say you should be testing your GI with no fewer than 1000 photons from the sun, upping it to anywhere up to 100k to 250k when optimised with some FG.

i am slightly confused by a few of the things you have described - firstly, you say that you fire 2 photons into your scene, this should just give you a few bright photon samples depending on bounces? changing the sampling radius at point 4 seems a little pointless as you dont have enough samples to begin with?

[SandmanNinja]

Hey Matt,

 

It's just a technique that I picked up from a video tutorial.

It's just a way to see where the photons are hitting. Are they energic enough? etc etc

 

Wow - never considered firing that many photons!

 

[time passes]

 

Okay, I set it to 100,000 and then 250,000 photons per sample.

It looked really good! I don't know how I missed that very important step the first time around. Guess I was too inexperienced to figure that out.

 

The blotches are all gone. But the area under the archways are actually darker here now. Hmm... If I remember, there are no multibounces when you use photons - is that correct? That would explain the area being darker as it is not in the direct sunlight.

 

Thanks again - be interested to get your feedback again.

 

Edited June 14, 2008 by SandmanNinja
added new render

[SandmanNinja]

And then the interior with more photons.

 

Still dark in places, but looking heaps better I think.

Still has some dark spots, but I can't do bounces with photons, right?

 

 

[odouble]

It's looking a lot better. I like your interior.

 

Do you have exposure checked in you environment settings? It should be if you are using the physical sky. You can adjust your exposure settings there to get a brighter scene without pumping more photons and upping your render times. You can also up your final gather bounces but at render cost.

[SandmanNinja]

Hi Tunde - thanks for the feedback and suggestions.

 

 

I tried to render the same scene, only with no photons/GI.

It took AGES, even with lower anti-alias settings.

 

I wasn't sure if values greater than 0 would be ignored or not when I used GI (I wrote down in my notes that you couldn't bounce in FG),

 

but I just did 2 quickie renders and compared them in RAM Player:

 

Render #1 (test_01.jpg):

800x600

GI; 250,000 photons; FG; Min: 1, Max: 4 - Bounces: 0; Time: 11 minutes 50 seconds

 

Render #2 (test_02.jpg):

GI; 250,000 photons; FG; Min: 1, Max: 4 - Bounces: 3; Time: 12 minutes 07 seconds

 

Ram Player and render time confirms that there is no difference between a scene with multiple bounces and a scene with 0 bounces, using GI.

 

I have attached my exposure settings - can you suggest something that I should change?.

 

Since I can't BOUNCE the photons in the render, I thought I'd try to fix the shadows in Post in Photoshop.

I went into Photoshop and did the following:

Image -> Adjustments -> Shadow / Highlight

I applied a 20% value to the shadow, brightening the shadows.

 

Thanks for the feedback - I appreciate it.

 

 

Scene with a background inserted via Alpha Channel and Image Adjusting the shadows

[Justin Hunt]

There are some basic concepts to be understood with photons

 

1) Photons are multi bounce, this is controled by the max trace depth in the photon settings. This number is the number of bounces, I usually use 30

 

2) FG is 0 bounce, if you set it higher, they are still calculated but then disregared at rendertime, open the message window and it will tell you as much. So leave FG at 0. The density, number of samples and intoplate are still relevant

 

3) Be sure to check the light photon settings, these will give you control for individual lights. Just switch between Automatic (which use the render setting numbers) and Maual, where you specify energy, number of photons and decay.

 

4) The physical scale in the exposure control panel is important as this sets the energy scale for the lights.

 

Overall Phons with FG is very quick. It does give different results than FG only rendering, Such as more colour variation in shadows. It may take a little time to get just the right numbers but its worth it in the end.

 

As to the darkening/more contrast. This is correct as with the more photons being shot, the more accurate the lighting solution will be. Another effect is more saturated colours

 

jhv

[SandmanNinja]

Justin,

 

THANK YOU very much!

 

I 'knew' that photons bounced, but I didn't know where to control how much they bounced until you showed me the light.

Tripling the number of bounces really got those photons in the nooks and crannies.

 

I know when I used Log (instead of mrPhotographic), I'd set the physical scale between 80,000 and 120,000 (as per Brian Bradley's tutorial) and it looked really good and proper. There doesn't seem to be a 'counterpart' for the physical scale in mrPhotographic...?

 

Render time went from 10min 50sec to 15min 23sec, and Ram Player shows the shadows corners aren't so dark with the 30 setting.

And the colours do look richer.

 

I've added your post to my notes.

[mattclinch]

I think you need to look at your photons again. 'photons per sample' is the number of photon hits to an area that it will approximate over when looking or the general light level. ie. '1 photon per sample' will place 1 GI sample at every photon bounce leading to a very very 'blotchy' render. the trick is to keep this value as low as possible to retain the tightest GI solution. the actual number of photons emitted into the scene is controlled by the 'average GI photons per light' value.

if you set the 'photons per sample' to 250000 it will just average each GI sample over the entire number of photons in the scene.

try emitting 100000 photons from your sun and averaging over 250 phtons per sample. optimised your GI for your FG (checkbox) and merge nearby sample to 0.0 so that two samples placed over each other will be discounted.

 

hope that helps.

[SandmanNinja]

Hey Matt,

 

Thanks for that.

 

I'll have a fiddle around and record some render times, compare output and note settings.

 

I think my nounaphasia is kicking in = the phrase "Photons per" appears in 3 different fields. Just gets a bit confusing.