Off Topic Discussion

View unanswered posts | View active topics


Post a new topicPost a reply
Previous topic | Next topic 

User avatar   Uber Modder Uber ModderMember since: 15.10.2009, 10:44Posts: 1603Status: Online Likes: 177
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 11.07.2012, 05:45 
Quote by Ultimarage:
Individual hairs on the characters were tessellated geometry using vector based procedural displacement. That's one thing CE3 can't do.

The object density is something CE3 can't do either. It looks like the base tri count is ten to twenty times of what is even possible in CE3 as well. Your limit in CE3 is still around 1.5 to 2 million triangles at 1080p, even on the most high-end hardware available... whereas the Final Fantasy scene had times where 20 million triangles were being displayed for base geometry, and that isn't even mentioning tessellated geometry.


20 mil tris for base geo? There are tons of low poly geo with normal maps. In some scenes you clearly see that. They are using instancing, «smart» texturing, proper light and geometry setup. It means you can use a lot of low poly objects but overall scene look will be much «higher» than it is. From certain you cant even tell is it high poly or its low poly.

CE3 limit is much higher than 2 mil tpf cuz i have some CE3 based scenes with more than 3 000 000 tpf and 40 fps on my middle end pc.

Quote by ClockworkOnion:
the engine also doesn't have any kind of GI solution pre-baked lighting, among other things.


We can use diffusely convolved cube maps and achieve GI. But because CE3 material editor is very limited we cant mix dccm and normal cm :(

Quote by ClockworkOnion:
while the next revision of CE3 and UE4 are probably going to leave it in the dust.


New CE3 revision is not so impressive. They added some nice features, but... its far away from what we expect. I hope they will add some more feature and next revision become much better than it is right now.

User avatar   Beginner BeginnerMember since: 13.05.2011, 15:56Posts: 36 Likes: 0
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 11.07.2012, 08:45 
Quote by Plastix:
New CE3 revision is not so impressive. They added some nice features, but... its far away from what we expect. I hope they will add some more feature and next revision become much better than it is right now.


I was caught by surprise with their apparently tessellation-free displacement mapping.
User avatar   Has no life Has no lifeMember since: 02.02.2007, 07:25Posts: 5071 Likes: 218
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 11.07.2012, 08:58 
I think it's just improved POM. Not displacement mapping.
Looks lke this...

Realtime Per-Pixel Displacement Mapping on Arbitrary Geometry
http://www.divideconcept.net/papers/RPPDM-RL04.pdf

Anyway, real displacement mapping need tessellation, it has real geometry.
If we use desplacement mapping on original low-poly model, it makes terrible result.
"No tessellation" means "Fake". But the fake is good in most cases.

I feel... we already saw this improved POM in latest CE3 SDK?


My Crytek News Blog - http://bit.ly/aFSbnP
My Massive Crysis Maps Guide - http://bit.ly/f5ifhe
My CryEngine Screens - http://bit.ly/wIDNDJ
My Youtube - http://bit.ly/xIg1l1
My Twitter - http://bit.ly/yZtaoH0
User avatar   Experienced Modder Experienced ModderMember since: 01.07.2011, 21:28Posts: 436 Likes: 0
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 11.07.2012, 22:38 
Quote by Staticman:
Quote by Plastix:
New CE3 revision is not so impressive. They added some nice features, but... its far away from what we expect. I hope they will add some more feature and next revision become much better than it is right now.


I was caught by surprise with their apparently tessellation-free displacement mapping.

you were surprised they started using POM? lol.
User avatar   Superstar Dolphin Superstar DolphinMember since: 16.08.2004, 10:46Posts: 9838Location: Finland Likes: 170
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 17.07.2012, 15:09 
They now basically have POM with no horrible distortion at grazing angles plus proper silhouettes. For most cases it'll look better than tessellation which will always have jagged edges depending on the size of the triangles. ;)
User avatar   Hardcore Modder Hardcore ModderMember since: 28.10.2007, 15:25Posts: 844Location: Denmark Likes: 1
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 17.07.2012, 15:21 
POM with better occlusion and proper silhouettes? So relief mapping essentially?

Also WRT jaggy tesselation, it's a simple and solved aliasing problem.
User avatar   Superstar Dolphin Superstar DolphinMember since: 16.08.2004, 10:46Posts: 9838Location: Finland Likes: 170
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 17.07.2012, 16:47 
Quote by BmB:
POM with better occlusion and proper silhouettes? So relief mapping essentially?

Relief mapping, cone step mapping or some other similar solution - Cry-Styves couldn't disclose any technical details when I asked. :)

I'm not sure what you mean by the aliasing problem being solved - I've read that article and didn't really notice anything related to this specific problem (other than the mention that triangles smaller than a few pixels are less effective to render).

This is what I mean:
Image
Image

Would be great if the new displacement mapping solution gets rid of the issues of POM at corners, while looking as good on flat surfaces. ;)
User avatar   Hardcore Modder Hardcore ModderMember since: 28.10.2007, 15:25Posts: 844Location: Denmark Likes: 1
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 17.07.2012, 18:02 
The point is jaggyness is due to aliasing, which is also what causes texture shimmering. Which makes sense since it's reading values from a heightmap texture. Aliasing is a solved problem.
User avatar   Superstar Dolphin Superstar DolphinMember since: 16.08.2004, 10:46Posts: 9838Location: Finland Likes: 170
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 17.07.2012, 20:09 
I don't see how the aliasing solution would solve this issue; it's not a shimmering artifact (stemming from sampling different heightfield points at different tessellation levels) but simply the result of the tessellated surface not aligning to the heightmap features and hence producing a zigzag effect on some straight edges. The features won't align even if you use a lower-res mip for the heightfield with low tessellation factors. It would just produce less accurate displacement (compared to the POM reference), even if it would reduce the jagginess. ;)

In order to get POM-accurate displacement you'd need at least one vertex per pixel (which is starting to weigh in on performance at least in the Cryengine 3 tessellation solution).
User avatar   Hardcore Modder Hardcore ModderMember since: 28.10.2007, 15:25Posts: 844Location: Denmark Likes: 1
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 17.07.2012, 22:34 
Well it sounds like you're not obeying the nyquist rate then.
User avatar   Uber Modder Uber ModderMember since: 26.11.2007, 02:03Posts: 2830Location: USA Likes: 117
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 18.07.2012, 07:23 
Grim is correct, this is not in fact an Aliasing issue. It is a limitation of the number of triangles that can be displayed. The new POM is a pretty interesting work around, and in terms of design is an amazing achievement. I've been thinking and trying out ways of doing this myself.
User avatar   Hardcore Modder Hardcore ModderMember since: 28.10.2007, 15:25Posts: 844Location: Denmark Likes: 1
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 18.07.2012, 12:03 
It absolutely is, you can see some of the edges are straight, which is where the vertices align with the heightmap texture, and those that are jagged is where they are misaligned and so sample from different heights. This is why the nyquist rate exists, because if you have 2x the vertices compared to texels, then they will always fall on a texel and smoothly interpolate. The other way to fix it is to make sure your vertices are perfectly aligned with your heightmap.

(and of course to downsample your heightmap)
User avatar   Superstar Dolphin Superstar DolphinMember since: 16.08.2004, 10:46Posts: 9838Location: Finland Likes: 170
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 18.07.2012, 14:16 
Quote by BmB:
It absolutely is, you can see some of the edges are straight, which is where the vertices align with the heightmap texture, and those that are jagged is where they are misaligned and so sample from different heights. This is why the nyquist rate exists, because if you have 2x the vertices compared to texels, then they will always fall on a texel and smoothly interpolate. The other way to fix it is to make sure your vertices are perfectly aligned with your heightmap.

(and of course to downsample your heightmap)

Yes that's exactly what I mean, you'd need much more dense grid of vertices to reach a POM-level quality. That much geometry starts to take its toll on your system at current implementations. Displacement works reasonably well for organic, natural forms even at low tessellation factors, but for hard, straight edges it's not the ideal solution because the artifacts start to be very visible.

If you try to fix the jagginess by making sure adjacent vertices sample the same height, the displacement won't be true to the heightmap, which again makes POM the better solution in that regard. Of course such a hard form would be easy to model in relatively few vertices, if the edges just align with the heightmap. ;)
User avatar   Hardcore Modder Hardcore ModderMember since: 28.10.2007, 15:25Posts: 844Location: Denmark Likes: 1
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 18.07.2012, 14:21 
I don't think you understand sampling here.
User avatar   Uber Modder Uber ModderMember since: 07.12.2007, 12:44Posts: 2410 Likes: 95
 

 Post Topic: Re: Agni's Philosophy - Realtime Tech Demo
PostPosted 18.07.2012, 21:09 
Those jagged lines are much larger than the possible triangle densities. I think BmB is correct - the triangle density should be able to get rid of those black waves no problem.

Look at the silhouette for the bricks on the edge of the wall. Perfectly straight.

I think this error just comes down to Crytek's problems with implementation. Tessellation and DX11 were slapped on as an after-thought.