Almost there with Lightmass!

[darkVoid]

Hello, I had to join because of a problem that i'm experiencing with Lightmass. Up until now I could figure out everything by googling and searching the forums, but this one is really giving me a headache. I hope someone here has a solution to this.

What i'm trying to do is increase the quality of the GI Lightmass produces. After searching the forums, I saw a moderator post where he/she mentions reducing the StaticLightingLevelScale value to achieve better GI. I'v done this, and these are my results.

[IMG] Info2 by Norbi905, on Flickr[/IMG]

I have one dominant directional light coming in through the windows on the right hand side of this image. I have absolutely no idea what those "dots" are. These seem to appear on both direct and indirect illumination parts. When I set the StaticLightingLevelScale back to default (1.0) they dissapear, but I also loose the quality of GI. Here is my WorldInfo Lightmass


Info by Norbi905, on Flickr

Regardless of that setup, I have tried to do a clean install of March2011 UDK. Created two boxes in max, with 2 UV sets. Exported into UDK and the only change I made within UDK was added a default white material. Added a point light and changed it's StaticLightingLevelScale to 0.05 and again this is the result I get. Same "dots" all over the image.


Info3 by Norbi905, on Flickr

I thought maybe not enough photons are being emitted, so I change the Lightmass .ini file to the highest (it seems like you can't emit more that 40mill) but to no affect. So I am out of ideas Any input is appreciated.

Thanks for reading this!

[percydaman]

there's a setting in the .ini somewhere that you can toggle compression for the lightmap. turn that off and its supposed to increase the quality, at the expense of package size.

[darkVoid]

I do apologize for the lack of information I posted earlier, it was a long night .

Compression is turned off within the lightmass ini file. I played around with Indirect Normal Influence Boost, but made no difference.

The texture size also doesn't make a difference. The screenshots provided were set to high 512 as well as 1024 lightmass texture but no noticable difference.

I tried UV both from within max (flatten mapping) and to have UDK generate one, none made a difference.

Sorry if I'v missed anything that might be of importance to solve this issue, please do let me know.

Thanks for the input percydaman, much appreciated!

[darkVoid]

A couple more screenshots to illustrate the problem. First is with StaticLightinLevelScale of 0.05, the second with default 1.0.


Info5 by Norbi905, on Flickr


Info4 by Norbi905, on Flickr

Since this is on preview build the difference is not that much, but you can still see that the GI is a lot better on the first one. Only if those spots would go away .

[horoxhoro]

the second with default 1.0 looks fine..try to boost indirect bounce more...to fill the dark area? my 2 cent

[darkVoid]

Thanks for the input horoxhoro. This is for an architectural visualization project and unfortunately detail is everything . I took another shot with StaticLightingLevelScale of 0.25 on production quality and as you can see there is a lot more detail which is good. However, the spots are still there


Info6 by Norbi905, on Flickr

[percydaman]

If your looking for a top notch quality arch vis render out of udk, then I think your going to need to look at prebaking some sort of lighting/occlusion into your textures. Or at the very least an occlusion map that you can multiply over your diffuse in your material. There was a thread awhile back that kinda got into getting good GI solutions for arch-vis, and frankly there didn't seem to be a really good solution. This is coming from someone with a pretty solid, professional background in architectural visualization, with years of experience in the field.

IMO, UDK, in its current iteration is kinda how GI renderers we're in their early stages. Can get pretty good splotch free solutions in exterior scenes, with alot of direct light and maybe one bounce of GI, but interiors, and their need for multiple secondary bounces, is where they start to stumble, with the splotches/noise.

Your screenshot above, looks pretty darn decent actually. It could be Brazil, before they included irradiance caching/mapping. Someone may correct me if I'm wrong, but I think UDK uses photon mapping in their GI engine, which is kinda old and busted technology at this point.

[horoxhoro]

yeaa im doing visualization too by using UDk..arch viz is have to be clean which is udk very hard to achieve that...if using too high resolution on lightmap its cos performance issues....

cheers good luck =)

[darkVoid]

Well I managed to remove all that "dirt/spots" that Lightmass was producing. I think I have reached the limit of what Lightmass can produce with the tools and documentation that we are provided with. There is still a few things i'd like to tweak, such as the corners of the walls, but this can wait. Thanks for all the input!


Untitled by Norbi905, on Flickr

[TheAgent]

Id be happy to know what lightmass settings were used? I can never get anything to look that good, as what you have looks great!

[percydaman]

yes, please share!

[crimson.cosmos]

Yes, please do!

cc

[thomas3711]

I joint the movement for sharing. Yes Please Mr. Share with us, your results are very very good

[darkVoid]

I do apologize for the delayed response, been extremely busy trying to get this product up and running.

It's kind of a bummer that there is really no way of controlling GI within UDK, other than adjusting StaticLightingLevelScale and of course the lighting build quality. In our case, we have a 64 node render farm, so build times are not really an issues. The quality of the lighting however is very important.

The only info I managed to gather was to reduce the StaticLightingLevelScale. So I did, instead of the default 1, I went to about 0.1 and even lower. By doing this, Lightmass did produce much better GI but at the cost of major artefacts that you can see from my previous post. Now i'm not an expert at GI, but I had a feeling this had to do with not enough sample being taken, or photons being emitted. By increasing NumHemisphereSampleScale, NumimportanceSearchPhotonsScale and NumDirectPhotonsScale the artefacts went away. I also doubled( at least) the values withing production quality build.

Sorry for the quick post, have to get back to work

[Misja Camfferman]

I've learned a lot from this...
also rendering my sunlight through window simulation with 0.25 lightingscaleresolution...
The results are great, It took me 8 minutes with 6 core AMD processor to calculated lightmasses...