Geometry vs Textures (Alien Isolation Art Question; Use Which?)

Sammy101

As you can see from the picture below I exported some assets from a art study that was done to replicate Alien Isolation art process; using geometry for many of the details, and without the need for so many normal maps. My question here is, which process would be better to do for the majority? Use vertex normals and more geometry, or have lots of normals for the texture process with high to low bake downs for environment assets? I don't really know which one to do as they both seem good. I'm looking for one that would be good for decent-great computer specs. Please help.

JackDCaron

Hey Sammy, IMO one thing CA did extremely well was use bump offsets in a lot of there textures. Things like ceiling panels with cone-like cavities with cylinders in the middle of the cavity, all bump offset. Pretty much a displacement map in conjunction with the camera angle that effects the position of each pixel in the color and information maps (as I understand it).

Not saying they weren't generous with the poly's, just something else to think about. If you don't care about limiting yourself, I would say getting the big to medium geometry reads in there and use your normal maps to create detail. Seems like instead of using straight normal maps for things like knobs and push-buttons they created geo, but nicks and folds were normal mapped.

Anyone who has seen this in the Rift can attest to how important having either geo or bump offset is. Normal maps really flatten out on the larger details save for super glossy materials.

marks

Yeah go nuts with the poly detail, we used geometry for soooooo much stuff in that game. Jack is pretty on the ball, there wasn't too much parallax mapping used though.

Angry Beaver

And from what I've heard normal maps don't sell well in 3D so using more geometry makes whatever you're doing more occulus ready.

Mant1k0re

Yeah normal maps are completely useless in VR.The illusion does not hold even a medium distance, it looks like splashes of paint.

Paralax works to some extent though.

JackDCaron

Marks, awesome, although it made me a bit sick (as most VR experiences do) the game really holds up in the headset incredibly well.

Surely there must have been some hi-res sculpts done for the seating and lived in spaces?

Sammy, I would say the worst that could happen is you have a bunch of geo that you could bake down if needs be, but reads really well if not.

Sammy101

JackDCaron wrote: »

Hey Sammy, IMO one thing CA did extremely well was use bump offsets in a lot of there textures. Things like ceiling panels with cone-like cavities with cylinders in the middle of the cavity, all bump offset. Pretty much a displacement map in conjunction with the camera angle that effects the position of each pixel in the color and information maps (as I understand it).

Not saying they weren't generous with the poly's, just something else to think about. If you don't care about limiting yourself, I would say getting the big to medium geometry reads in there and use your normal maps to create detail. Seems like instead of using straight normal maps for things like knobs and push-buttons they created geo, but nicks and folds were normal mapped.

Anyone who has seen this in the Rift can attest to how important having either geo or bump offset is. Normal maps really flatten out on the larger details save for super glossy materials.

Hey thanks Jack, I'll definitely be looking into the stuff you're describing, sounds tricky though. Another question I have is if the geometry method would have a bad effect if used in multiplayer. I wouldn't think so right, though I have no idea. Thanks for all the input guys .

Sammy101

JackDCaron wrote: »

Marks, awesome, although it made me a bit sick (as most VR experiences do) the game really holds up in the headset incredibly well.

Surely there must have been some hi-res sculpts done for the seating and lived in spaces?

Sammy, I would say the worst that could happen is you have a bunch of geo that you could bake down if needs be, but reads really well if not.

Hey Jack this is what you are talking about right? Parralax Mapping video from Youtube, I think it's what you're talking about.

[ame]https://www.youtube.com/watch?v=P-5Qzi-ljsc[/ame]

DireWolf

So bump parralax is sort of like Displacement?

Mant1k0re

Sammy101 wrote: »

Hey Jack this is what you are talking about right? Parralax Mapping video from Youtube, I think it's what you're talking about.

https://www.youtube.com/watch?v=P-5Qzi-ljsc

That's parallax indeed. I'm not sure why he says "bump offset map" instead of height map for the third input. It's the first time I ever hear this term...

But admitedly I'm very new to all this.

Sammy101

Mant1k0re wrote: »

That's parallax indeed. I'm not sure why he says "bump offset map" instead of height map for the third input. It's the first time I ever hear this term...

But admitedly I'm very new to all this.

So a bump offset map is just a displacement map?

dzibarik

is there any way to get rid of diffuse/albedo distortion in parallax?

Mant1k0re

Sammy101 wrote: »

So a bump offset map is just a displacement map?

I wish people worked out naming conventions for input maps! I believe the terms displacement maps and height map are sometime used interchangeably, but I'm not sure they actually are because it's actually 2 separate outputs in Allegorhytmics' Bitmap2Material for instance. Crazy Bump will just spit out a displacement map on the other hand.

They're both greyscale maps in any case.

Now if someone more experienced could chime in! I'm also insterested in knowing if it's just semantics.

JackDCaron

Sammy, that's it! Parallax == Bump offset. I believe he calls it a bump offset map for cohesion within Unreal, but it's just a height map. It distorts the pixels of the map to create a sense of depth without actually moving any verts like a true displacement would.

Height maps and Displacement maps are basically the same. They are both intended to move vertices. A vector displacement has the same function, but is more directional, like a normal map versus bump map.

dzibarik - Because it is a fake it has limits. Just play with the height setting (a little goes far) and maybe make sure you aren't too drastic with the falloff from low to high.

Sammy101

JackDCaron wrote: »

Sammy, that's it! Parallax == Bump offset. I believe he calls it a bump offset map for cohesion within Unreal, but it's just a height map. It distorts the pixels of the map to create a sense of depth without actually moving any verts like a true displacement would.

Height maps and Displacement maps are basically the same. They are both intended to move vertices. A vector displacement has the same function, but is more directional, like a normal map versus bump map.

dzibarik - Because it is a fake it has limits. Just play with the height setting (a little goes far) and maybe make sure you aren't too drastic with the falloff from low to high.

Hey Jack thanks for the reply; That's good to know though. I will follow this video, looks a lot better than just normal map alone.

marks

Bear in mind that parallax mapping is pretty expensive, we actually disabled it on all of the shaders for PS3 and 360 on A:I because it was so expensive to render.

Sammy101

marks wrote: »

Bear in mind that parallax mapping is pretty expensive, we actually disabled it on all of the shaders for PS3 and 360 on A:I because it was so expensive to render.

Hey Marks, may I ask how expensive it actually is though? Should I be able to use it on the majority of textures that would benefit of it on PC? Or should I use it sparingly?

Farfarer

Depending on the implementation, it's pretty expensive. You can wind up with tens of samples per pixel...

JackDCaron

Hmmm, good to know. I've only used them on hero elements in a couple VR projects. They probably live somewhere between additional geo and tessellation in terms of performance, but definitely makes the shader heavier.

Sammy101

Update: I figured out how to manually set vertex normals, quite interesting and cool. However setting them up manually takes a long time. Isn't there some way to speed this up heh.

ZacD

Are normal maps really completely useless in VR? I think they'd still work as detail maps or for micro detail, like the subtle texture on wall paint. They aren't going to make a 6 sided cube look like a rounded box anymore, but when there's detail that is too small to model or use with tessellation, it might still be better than nothing.

Also people don't get any depth perception for things really far away in the distance, normal maps should be able to be used there as well.

Chev

Mant1k0re wrote: »

I wish people worked out naming conventions for input maps! I believe the terms displacement maps and height map are sometime used interchangeably, but I'm not sure they actually are because it's actually 2 separate outputs in Allegorhytmics' Bitmap2Material for instance. Crazy Bump will just spit out a displacement map on the other hand.

A heightmap's originally a specific case of displacement map used for landscapes, the one where all displacement is vertical based on a flat or spherical original terrain (thus heightmap, ie each texel specifies a terrain height). Displacement map are the general case with an arbitrary mesh and displacement along normals.

FelixL

ZacD wrote: »

Are normal maps really completely useless in VR? I think they'd still work as detail maps or for micro detail, like the subtle texture on wall paint. They aren't going to make a 6 sided cube look like a rounded box anymore, but when there's detail that is too small to model or use with tessellation, it might still be better than nothing.

Also people don't get any depth perception for things really far away in the distance, normal maps should be able to be used there as well.

I've done some VR stuff and I don't think they are useless.
Of course, you can't do stuff like a flat wall with normal mapped pipes and all sorts of crazy protruding detail in a normal map only.
Also, you can start worrying about fine detail once the displays are 4k :P In my experience, it's hardly worth using textures larger than 512 because the resolution is so bad. Also, the pentile doesn't help.

Sammy101

ESF wrote: »

Sammy101: looking nice! and no i'm not using any scripts but in maya in the bevel settings you can control the smoothing

Hey guys so I talked with ESF and he said he could control the smoothing Maya through Bevel Settings. I have no idea what he's talking about, but if I could use my student version of Maya to control the vertex normals and then export, that'd be awesome. Unfortunately like I said I have no clue on what he's talking about, so I'm hoping someone can fill me in.

cptSwing

marks wrote: »

Yeah go nuts with the poly detail, we used geometry for soooooo much stuff in that game. Jack is pretty on the ball, there wasn't too much parallax mapping used though.

Off topic really, but congrats on such a purdy game. captured the vibe of Alien so, so well. (just got it recently)

Mant1k0re

ZacD wrote: »

Are normal maps really completely useless in VR? I think they'd still work as detail maps or for micro detail, like the subtle texture on wall paint. They aren't going to make a 6 sided cube look like a rounded box anymore, but when there's detail that is too small to model or use with tessellation, it might still be better than nothing.

Also people don't get any depth perception for things really far away in the distance, normal maps should be able to be used there as well.

You mean like high frequency details? Those are the worst imo -.-

Don't take my word for it though

OVR thread

Monster (developping for Playful, making one of the first game officially sponsored by OVR)

Chev wrote: »

A heightmap's originally a specific case of displacement map used for landscapes, the one where all displacement is vertical based on a flat or spherical original terrain (thus heightmap, ie each texel specifies a terrain height). Displacement map are the general case with an arbitrary mesh and displacement along normals.

Thank you for taking the time to explain this. It's much appreciated (and noted down).

Joost

If you use max you can use quad chamfer to automatically change vertex normals around the chamfers [ame]https://www.youtube.com/watch?v=34giO1tu43M[/ame]

Sammy101

Joost wrote: »

If you use max you can use quad chamfer to automatically change vertex normals around the chamfers https://www.youtube.com/watch?v=34giO1tu43M

Hey Joost! Thanks for the reply, that's very interesting and I am definitely going to check that out. I also today though, finally got the solution to my problem. I talked with Liam and he was amazing help and never would have figured it out without him, I got the knowledge on how to properly chamfer and set vertex normals through smoothing groups so I can have objects appear to have normals without normal maps.

edit: Hey Joost, I checked my 3DS Max and I couldn't find Quad Chamfer. Is it a maxscript? I am using 2012 however, so perhaps I'm outdated.
edit2: I just realized you were the guy who made those awesome marmoset skies. Amazing work man, truly awesome.

Mant1k0re

Sammy101 wrote: »

edit: Hey Joost, I checked my 3DS Max and I couldn't find Quad Chamfer. Is it a maxscript? I am using 2012 however, so perhaps I'm outdated.

Yes, it's a Max 2015 feature.

Joost

Actually I don't think max 2015 has the vertex normals feature in the integrated version of quad chamfer. It's a plugin you can buy here: http://www.mariussilaghi.com/products/quad-chamfer-modifier

Bit pricey but worth the money imo.

Setting vertex normals through smoothing groups sounds interesting, is that through a script?

Sammy101 wrote: »

edit2: I just realized you were the guy who made those awesome marmoset skies. Amazing work man, truly awesome.

Thanks!

marks

Info here: http://wiki.polycount.com/wiki/VertexNormal

On that wiki page there is the "GetVertNormalsFromFace" Maxscript available to download. That's what we started off using on A:I, but we ended up writing our own vertex normals toolset which was used for most of the project.

Pretty sure this technique was used on like 80-90%+ of environment geometry in the game.

Joost

Syncviews script works well but it's such a tedious process imo. Quad chamfer is parametric so it saves quite a lot of time. Though I haven't done any extensive testing with the quad chamfer method.

edit: This script is quite useful too, but it requires chamfers everywhere because it ignores smoothing groups. http://www.polycount.com/forum/showpost.php?p=2230470&postcount=28

Sammy101

Joost wrote: »

Syncviews script works well but it's such a tedious process imo. Quad chamfer is parametric so it saves quite a lot of time. Though I haven't done any extensive testing with the quad chamfer method.

edit: This script is quite useful too, but it requires chamfers everywhere because it ignores smoothing groups. http://www.polycount.com/forum/showpost.php?p=2230470&postcount=28

Hey Joost just checking back in the thread and I watched a few videos of quad chamfer online. Do you have to set smoothing groups up for the quad chamfer to work with vertex normals? Or is it somehow done through the modifier?

edit: In this video in the very beginning, is he editing the vertex normals through the modifier here? [ame]https://www.youtube.com/watch?v=jMgd7RrWgww[/ame]

Joost

You can use smoothing groups for areas where you don't want chamfers but you don't need to set them for the vertex normals. The modifier changes the normals around your chamfers automatically and it should preserve your smoothing groups. Just set the shading to sharp boundary with 1 round shade. And then do a chamfer by edge weight (or smoothing group.)