Sample Distibution Shadow Maps


photo

Recommended Posts

New Sample Distribution Shadow Map algorithm shows impressive shadow map quality due to adaptive shadow split calculation. Homepage includes paper, movie and DirectX 11 source code. Movie demonstrates dramatic better quality compared to PSSM (no visible split edges and shadow undersampling) especially during outdoor scene zoom in/out.

 

Would be great to see this in UNIGINE because current PSSM shadow implementation cannot handle such kind of scene renderings with largely varying camera z-coordinates (e.g. architectural fly-throughs, birds-eye scene renderings, flight simulations etc) without mentioned artifacts.

Link to post

this is greate shadow tech!! I hope to see this in unigine :( +1

Link to post
  • 4 months later...

@UNIGINE: any chance of getting this into Unigine. Described technique seems to produce so much better shadow quality !

Link to post
  • 1 month later...

Thank you for these links! However, it seems this approach is more demanding on GPU, so higher quality would need to be paid for with decreased performance, which is hardly acceptable.

Link to post

Hi manguste,

 

actually the Sample Distribution Shadow Map paper claims HIGHER performance in most cases compared to static PSSM while providing dramatic improved shadow quality (have a look on the videos).

 

Current UNIGINE PSSM implementation also suffers from quite visible shadow seams between shadow splits, especially if camera view changes very dynamically (e.g. camera fly-throughs). Also current implementation does not provide any kind of smooth fading for the last shadow split (see this post) which causes hard shadow/non-shadow borders extremly visible on high-flying camera positions (e.g. birds-eye-views)

 

At least for our virtual reality customers quality is absolutely paramount and visual artifacts are always a shop-stopper while GPU power increases on a montly scale...

Link to post

manguste, let me quote a phrase from first link:

 

Finally, we demonstrate that SDSMs outperform PSSMs in a large 2009 game scene at high resolutions, making them suitable for games and other interactive applications.

 

Also I'm not sure that Unigine have higher GPU requirements than CryEngine 3 and even if so, it is not a reason to avoid new technologies..

Link to post

At least it would be great if frustum could have a look on SDSMs for an estimation of technical feasibility. Implementation might be straight forward (above link also provides full source/shader code) while quality improvements and ease of use would be dramatic. UNIGINE should stay on top of rendering quality as demonstrated in the past. Current shadow implementation seems a litte bit outdated compared to other top-notch engines

Link to post

tested with the sample program, with sdsm, the FPS is about 130(default setting), but pssm's fps is 180, pssm is faster, but has so many visual artifacts.

 

SDSM has best visual effects.

Link to post

There are several downsides of such dynamic shadow map focusing techniques:

* The density of shadow map pixels is dynamic and because of that we will have bad looking shadows sometime.

* The softness of shadow will be dynamic, so it even worse than the first drawback because soft shadows is not supported in this case.

Link to post

We are very much in the same boat as Ulf. We do a lot of birds eye stuff . We do a lot of work making the current implementation work for us but any improvements to the visual quality of the shadows would be very welcome.

Link to post

Danny, that's exactly the point. Actually it doesn't matter if it's PSSM, VSM, SDSM,... or what ever SM. Especially for out-door-/ birds-view-scenes with long view distances current shadow implementation is more or less inapplicable due to shadow seams and missing distance fading...for these use-cases other engines provide much better solutions :)

Link to post
  • 1 year later...

Hi,

 

We would like to bring this topic of shadow quality up again.

 

We still struggle to hide the very noticeable visual quality of shadow seams between splits. Every high level predefined camera path has this problem.

 

It really ruins a good thing.

 

Thanks,

N.

Link to post

Unfortunately UNIGINE crew ignores this flaw and all customer complains since years....

Link to post

It really ruins a good thing.

 

 

 

I think that this is THE point with every flaw.

95% of a project can be excellent and 5% can ruin everything or even one tiny flaw.

 

It is not about the crititc of the audience, most of the time it is definitely disturbing and irritating, and the critic follws these issues.

 

For our projects, we fight very hard against these things as they are (and for us this is what it says) a showstopper.

 

I did' nt come over this shadow problem for now, as I am starting with Unigine (I just saw this issue in the Valley demo, where imo it is VERY annoying), but I hope that, and it seems really crucial, treated seriously.

 

Best.

 

Werner

Link to post

We have projects with issues raised Internally about shadow flickering and quality for a long time also, however it is only fixable by Unigine so these project bugs still remain open and i expect customers to start complaining when these projects are delivered soon.

Link to post
  • 4 weeks later...

We just loaded an old level into the latest unigine version. The SSAO is looking great in fact mostly everything is looking great except the shadows. What can be done about the shadows?

Link to post
  • 4 weeks later...

Blending between shadow map splits will hurt performance with PCF filtering enabled. Because in that case we have to perform two expensive shadow filtration kernels. We will improve the shadows quality but I can't say you when it will be implemented.

Link to post

too bad news, long-distance shadows with smooth blending are really missing

Link to post
  • 3 months later...