I've heard from sources that you should NOT let your normal maps do a lot of heavy lifting. If you can actually model the depth of an object, do it, otherwise it's gonna look worse in VR.
As far as I can tell, it's only those considerations that really change within a VR environment.
I heard something about transparency taking more performance to render in VR. (Don't take my word for it, if anyone can verify this it would be appreciated.)
Detail maps are something you should research. It allows you to have lower resolution textures and still have some nice details. Optimization is super important since the game needs to run at a set FPS, whether it's 90 or 120. Thinking about re-usability is also important as well as combining texture sets if possible.
They are suffering in terms of resolution. A friend of mine has tried out Robinson: The Journey and he said it looks alright up until you get close. Overdraw is still and issue, but it can be somewhat minimized by smart branch texture layout then modeling the lowpoly branches as close to the texture as possible leaving minimal transparency.
Been trying to confirm in VR the poor showing reliance on normals gives... But sadly all I got is gearVR at the moment. Got to admit... Pipelines that start phasing out normals sounds like a wet dream to me! Though I really doubt we are at the point where we can just throw our hi freq zbrush geometry anytime soon. I could see stylistically using a medium high geometry detail pipeline and taking a vacation from normals.
Been trying to confirm in VR the poor showing reliance on normals gives... But sadly all I got is gearVR at the moment. Got to admit...
Normals still work pretty well in VR (from personal experience). As Brian said, it's about not having them do a ton of heavy lifting. I think the whole no normals thing was/is blown out of proportion a bit.
I think as more artists get their hands on VR kits they will be able to figure out what works and what doesn't. I'm speaking from a weapons perspective but I'm sure environments will have their own limitations and workarounds.
The no normal map thing is basically normal maps are not a substitute for actual depth, you still need normal maps for lighting to look right on small scale details. You can't slap a decal of a vent made in NDO and call it good (which games don't really get away with as much anymore anyway).
Right I would love to see for myself... perhaps compare with tessellation stategies? But Nvidia's Tom Peterson sort of sounded perty adamant on normals being obsolete in VR where the lighting model does not hold up and looks fake? ( from the pcper interview ): not sure why the vid isnt starting at 17m:12s? but u can simply scroll to view to the normals in VR discussion. https://youtu.be/xtely2GDxhU?t=17m12s
To get a sense of the way normalmap sometimes work and sometimes don't in VR ... simply look at the realtime model for the Vive controller. It is the typical scenario when normalmaps would absolutely fail and look downright ugly, because of the amount of scrutiny this smooth and detailled object needs to hold up to. It's hard to explain, but it becomes immediately clear once you put on the HMD and inspect the controller up close - A 2048*2048 or even 4096*4096 normalmap bake would probably still reveal pixels.
As for normalmaps in VR in general : after testing it out I'd say they can still look great on things like a highres character model. But I can definitely imagine quite a few cases when they would look cheap. In my opinion there is currently no game available in VR that can fully put a lid on the question, but everything will probably be more clear once the DoomVR game/demo comes out.
@pior I have to think, with the minimum system requirements for the VR PCs, we should start to get more comfortable with medium density models, eh? And leave the normal maps for mostly higher noise details.
It's really interesting because it seems to vary so much depending on context. As said the Vive controller definitely required dense geometry, which in turn seems to indicate that gun models that can be inspected up close will have to be done that way too ; but still, Dota2 characters (which in theory shouldn't look to hot in VR because they are lowpoly, lowres and normalmapped) look fantastic, even when viewed at human scale. Maybe the low resolution of current VR HMDs is yet another factor to consider ...
Distance is a huge factor, the further away something is, the less depth we get from our eye sight. But we still get some parallax from moving around. But for something in our hands or right in our face, depth really needs to be supported by geometry. Detail normals are definitely important, you can't get paint/fabric/leather texture with tessellation.
pior raises an interesting point... where charming volume might be inherently "stereoscopically charming". I have noticed as much in 3dvision. Where complex low poly worlds have a stereoscopic charm not necessarily appreciable
when viewed normally. ( abnormally monoscopic )
Noticing and enjoying as much... It has always bothered me that 3Dvision development does not really seem to be of concern at dcc asset creation phases of development when viewing sculpts in 3d at the very beginning can expose that stereoscopic charm in volume and silhouette! And one has to buy a quadro card in Maya to adopt NVidia's 3dvision "Pro" line of active shutter expense. I hope more stereoscopic methods are now exposed during development like with Unreal's VR editor tools! So stereoscopic consideration is no longer an afterthought.
Distance is a huge factor, the further away something is, the less depth we get from our eye sight. But we still get some parallax from moving around. But for something in our hands or right in our face, depth really needs to be supported by geometry. Detail normals are definitely important, you can't get paint/fabric/leather texture with tessellation.
Right Detail normal/bump maps do not fall apart as you move around a surface's alignment Like Petersen described in the video. But he even begins by explicitly naming bump maps! ( instead of an obvious choice like relief ). But then again he creates a penalty jar in the video for every time he names Maxwell when he means to say Pascal! I would hope that one of NVidia's prominent engineers would not have me second guessing myself at this early stage of trying to build a new medium's language. But I think we can rest assured that your right and it was a slip. Would sure save a lot of time and headache if I could take a nice relaxing vacation from baking concerns and return to the ease where even just a greyscale bump map may be "good enough".
Been doing a lot of VR concerns research lately... Would be nice to see more threads pop up now that things are starting to accelerate. Really interested in other's concerns/findings/strategies?
Replies
As far as I can tell, it's only those considerations that really change within a VR environment.
It is verified. I've seen people use dithering as the work around, especially given the resolution of the current OLED screens.
How viable are heavy foliage scenes such as forests in VR right now?
But sadly all I got is gearVR at the moment.
Got to admit...
Pipelines that start phasing out normals sounds like a wet dream to me!
Though I really doubt we are at the point where we can just throw our hi freq zbrush geometry anytime soon.
I could see stylistically using a medium high geometry detail pipeline and taking a vacation from normals.
I think as more artists get their hands on VR kits they will be able to figure out what works and what doesn't. I'm speaking from a weapons perspective but I'm sure environments will have their own limitations and workarounds.
perhaps compare with tessellation stategies?
But Nvidia's Tom Peterson sort of sounded perty adamant on normals being obsolete in VR where the lighting model does not hold up and looks fake? ( from the pcper interview ):
not sure why the vid isnt starting at 17m:12s?
but u can simply scroll to view to the normals in VR discussion.
As for normalmaps in VR in general : after testing it out I'd say they can still look great on things like a highres character model. But I can definitely imagine quite a few cases when they would look cheap. In my opinion there is currently no game available in VR that can fully put a lid on the question, but everything will probably be more clear once the DoomVR game/demo comes out.
I have to think, with the minimum system requirements for the VR PCs, we should start to get more comfortable with medium density models, eh? And leave the normal maps for mostly higher noise details.
pior raises an interesting point...
where charming volume might be inherently "stereoscopically charming".
I have noticed as much in 3dvision. Where complex low poly worlds have a stereoscopic charm not necessarily appreciable
when viewed normally. ( abnormally monoscopic )
Noticing and enjoying as much...
It has always bothered me that 3Dvision development does not really seem to be of concern at dcc asset creation phases of development when viewing sculpts in 3d at the very beginning can expose that stereoscopic charm in volume and silhouette! And one has to buy a quadro card in Maya to adopt NVidia's 3dvision "Pro" line of active shutter expense. I hope more stereoscopic methods are now exposed during development like with Unreal's VR editor tools!
So stereoscopic consideration is no longer an afterthought.
Right Detail normal/bump maps do not fall apart as you move around a surface's alignment Like Petersen described in the video. But he even begins by explicitly naming bump maps! ( instead of an obvious choice like relief ).
But then again he creates a penalty jar in the video for every time he names Maxwell when he means to say Pascal!
I would hope that one of NVidia's prominent engineers would not have me second guessing myself at this early stage of trying to build a new medium's language. But I think we can rest assured that your right and it was a slip.
Would sure save a lot of time and headache if I could take a nice relaxing vacation from baking concerns and return to the ease where even just a greyscale bump map may be "good enough".
Been doing a lot of VR concerns research lately...
Would be nice to see more threads pop up now that things are starting to accelerate. Really interested in other's concerns/findings/strategies?