Ask for AA improvements

[lin.xianming]

Hi,

 

There are two anti-aliasing algorithms in Unigine, FXAA and TAA. 

Since our project is related with video broadcast industry, and our customers are really picky. 

So,Personally,I don't think the quality of FXAA is good enouth and acceptable.

As to TAA, It looks much better than FXAA，but It relies on a couple of previous frames, so it causes two disadvantages:

1>The outcome seems a bit vague, especially when we take video as a element and make it play a very important role, the contents of the video become vaguer obviously. 

2>When we change the viewpoint from one camera to another, the output is not as clear-cut as it's supposed to be.  

 

TAA is OK for games, but definitely not for video broadcast apps.

Is it possible that Unigine provides a better AA algorithm? 

 

Thanks!

[silent]

lin.xianming,

 

Unfortunately, no other AA technique can give us that anti-aliasing quality level / performance ratio which TAA gives us currently. If you are not satisfied with TAA you can try even to disable it and use supersampling rendering (render_supersampling console command) to get the anti-aliased image as a result. You can also try to use SLI / Crossfire to get more performance with TAA off. 

 

We are also working at MSAA implementation, but the performance drop is huge here. However, the SLI / Crossfire is working well with SLI / Crossfire, so you can spend more on hardware and get a good results.

 

Thanks!

[lin.xianming]

Hi Silent,

 

SSAA also causes vague, so it's out of place.

 

MSAA is good. Just the edges of the objects are processed, It can maintain the sharp of the textures. But I got an idea that the hardware MSAA can't get along with deferred rendering. I hope I'm wrong.

I try to modify the sample "main_sdl", and make it render some objects out, It works. then I change the parameter "SampleDesc.Count"of "\source\app\main_sdl\AppSDL.cpp", and intend to implement MSAA, But the outcome becomes black unless "SampleDesc.Count = 1".

 

Could you please give me some details about how to implement MSAA in Unigine?

 

Thanks!

[silent]

lin.xianming

 

A part of internal rendering logic should be refactored in order to get MSAA done. I'm not sure that it's currently possible without engine sources. We are already started research on this and maybe in 2.6 we can provide working MSAA for deferred rendering pipeline.

 

For the current moment you can just use supersampling without TAA.

 

Thanks!