An Outsiders Perspective on the OpenDungeons Media

An Outsiders Perspective on the OpenDungeons Media

Postby p0ss » 06 Mar 2011, 03:38

So yesterday I went through your entire media directories, both production and source, I've looked at every file and I took notes on my conclusions. These are my recommendations:


Models:

Firstly, pretty much everything is extremely low poly, I can only assume it is due to the skill level of contributors as it certainly isn't winning you much in terms of performance. Polys are an outdated measurement for performance and say almost nothing about the in-game performance, vertex count is closer, but even still, modern hardware is at such a point that thousands of vertexes aren't really a problem any more. The vast majority of your performance is not dictated by vertex count, and I seriously doubt, given your game style, that you will ever see the tens of millions of vertexes required to produce any significant performance hit. My conclusion: Do away with low poly entirely, it looks crap and doesn't help the game, start using high poly, well optomised models and you will see what I mean immediate.y However, that doesn't mean there aren't gains to be had from your models:

The biggest performance hit from your models currently is that every single one of them has double sided faces, and many of them also have hidden faces. This is an absolutely no no. Double sided faces means that your game engine is drawing the inside of every single model in the game, effectively doubling the amount of system resources used to draw any given scene. Simply going through and making them single sided will give you a better performance increase than any amount of poly count hoarding. The hidden faces are also a big problem it means you are rendering things that no one will ever see, and that is a needless waste of resources, particularly when you are rendering both sides of those hidden faces.

Textures:

so you have these tiny poly models, and yet you give them gigantic textures? Is this some kind of over compensation? 2048x2048 png 24 for creatures that have a handful of faces and spend most of their lives taking up a tiny fraction of the screen is insane. You could redo all of your textures at 512x512 png 8 and barely even notice a difference.

Here is an image to demonstrate what I am saying, this is the troll texture, displayed at a level of zoom which would correspond to zooming right in on him ingame.

Image

the current Troll texture is on the right, 2048x2048 png 24, filesize 6,500kb
An alternate troll texture on the left, 512x512 png 8 filesize 136kb

That means your current troll texture is 47 times(!) larger than it needs to be. And on top of that it is being drawn twice by the engine, meaning your engine is doing 94 times more work than it needs to every single time it draws a troll.

There is another issue with your creature textures, in that many of them use mirroring in their UV unwrapping, which renders very poorly, and the situation only gets worse once you start using normal maps, since simply mirroring a normal map produces the wrong results, meaning your creatures end up looking blurry and the light bounces off them incorrectly. I'd recomend re-unwrapping and retexturing all textures that use mirroring, otherwise you are shooting yourself in the foot.

Normal/specular maps
These are sorely missing, I understand they are on the todo list, and I would make them a priority, particularly for your walls. If you are going to have low poly models then normal maps are of critical importance to avoid everything looking like boxes.

Environmental Textures
The number one thing missing from your engine at the moment is environmental texture variety, your whole world is made out of less than ten textures, all of which are tiny little tiles. There is simply no way to make that look good.

In this image I replaced the existing dirt and rock textures with higher quality versions,
Image
and then zoomed right in. As you can see that looks better, but it doesn't matter how good I make the textures, because each of those textures is then tiled hundreds of times in every direction. It is simply not possible to have any texture look good under those circumstances. What the engine needs is a way to provide texture variation. The two easiest ways to achieve this are:

slightly altered tiles
this is the easiest way, I just make up a set of dirt textures which are still seamless with each other at the edges but are different in the center.
pros: easy to implement
cons: still looks relatively crap as different texture types still jar.

Automatically Blending tiles.
pros: looks great, is easy to add new textures into and allows for huge variation in theme.
cons: difficult to implement.

Personally I'd be recommending the second option, but either way is better than what you have now.

a final point on environments, your project is meant to be dark, and yet most of your environment textures, and indeed many of your creature textures are lightly coloured. If you want to convey a feeling of darkness, shiny beige floor tiles are not the way to do so.


Folder Arrangement

My last point is that the contents of your folders need to be better arranged. At the moment, with a relatively small amount of media, its not too bad, but as you gain more an more content the situation is just going to get worse and worse. Having all of your environment textures, creature textures, item textures, material settings and effects, all in one folder... that is a recipe for utter chaos. I would highly recommend rearranging your folder hierarchy so things are more logically ordered and divided.

I mean look at this, its a mess:
Image

Final Conclusions

It seems to me that up until this point the art in the OpenDungeons Project has been functioning in a fairly ad hoc way, I would suggest that there needs to be some strategic direction and a unified artistic vision, which can best be achieved by establishing content guidelines and todo/request lists. This appears to me to be a key priority if OpenDungeons is ever to move form being a techdemo into being an actual game.

In regards to content guidelines, I'd suggest having a look at the DungeonHack content creation guidelines as a starting point: http://dungeonhack.sourceforge.net/Art_Workflow_3D

I hope I haven't trod on too many toes, and if I have I hope it doesn't hurt too much because I feel much of this needs to be said. Goodluck and good day.
p0ss
 
Posts: 26
Joined: 18 Jan 2011, 02:10

Re: An Outsiders Perspective on the OpenDungeons Media

Postby andrewj » 06 Mar 2011, 09:47

p0ss {l Wrote}:And on top of that it is being drawn twice by the engine, meaning your engine is doing 94 times more work than it needs to every single time it draws a troll.

You are very wrong there, rendering performance is not proportional to the size of the textures being used.
User avatar
andrewj
 
Posts: 194
Joined: 15 Dec 2009, 16:32
Location: Tasmania

Re: An Outsiders Perspective on the OpenDungeons Media

Postby svenskmand » 06 Mar 2011, 12:01

p0ss {l Wrote}:So yesterday I went through your entire media directories, both production and source, I've looked at every file and I took notes on my conclusions. These are my recommendations:


Models:

Firstly, pretty much everything is extremely low poly, I can only assume it is due to the skill level of contributors as it certainly isn't winning you much in terms of performance. Polys are an outdated measurement for performance and say almost nothing about the in-game performance, vertex count is closer, but even still, modern hardware is at such a point that thousands of vertexes aren't really a problem any more. The vast majority of your performance is not dictated by vertex count, and I seriously doubt, given your game style, that you will ever see the tens of millions of vertexes required to produce any significant performance hit. My conclusion: Do away with low poly entirely, it looks crap and doesn't help the game, start using high poly, well optomised models and you will see what I mean immediate.y However, that doesn't mean there aren't gains to be had from your models:

The biggest performance hit from your models currently is that every single one of them has double sided faces, and many of them also have hidden faces. This is an absolutely no no. Double sided faces means that your game engine is drawing the inside of every single model in the game, effectively doubling the amount of system resources used to draw any given scene. Simply going through and making them single sided will give you a better performance increase than any amount of poly count hoarding. The hidden faces are also a big problem it means you are rendering things that no one will ever see, and that is a needless waste of resources, particularly when you are rendering both sides of those hidden faces.

Yes this seems a bit stupid. Wonder how noticeable the speedup will be :)
p0ss {l Wrote}: Textures:

so you have these tiny poly models, and yet you give them gigantic textures? Is this some kind of over compensation? 2048x2048 png 24 for creatures that have a handful of faces and spend most of their lives taking up a tiny fraction of the screen is insane. You could redo all of your textures at 512x512 png 8 and barely even notice a difference.

As far as I understand we have them at these sizes because they are the souce texture, meaning that we can then resize them to smaller sizes for the game. We have not made any testing of how notice able the difference is yet, but that is surely something that needs to be done when optimizing the game.
p0ss {l Wrote}: Final Conclusions

It seems to me that up until this point the art in the OpenDungeons Project has been functioning in a fairly ad hoc way, I would suggest that there needs to be some strategic direction and a unified artistic vision, which can best be achieved by establishing content guidelines and todo/request lists. This appears to me to be a key priority if OpenDungeons is ever to move form being a techdemo into being an actual game.

In regards to content guidelines, I'd suggest having a look at the DungeonHack content creation guidelines as a starting point: http://dungeonhack.sourceforge.net/Art_Workflow_3D

I hope I haven't trod on too many toes, and if I have I hope it doesn't hurt too much because I feel much of this needs to be said. Goodluck and good day.

This is very true indeed, as I see it. But we have recently had good progress on styling guidelines for the creatures. We have been working on the style guidlines for the constructs and that work seems very promising.
Jamendo.com - The best music store on the net, uses CC licenses.
User avatar
svenskmand
OD Moderator
 
Posts: 1850
Joined: 09 Dec 2009, 00:07
Location: Denmark

Re: An Outsiders Perspective on the OpenDungeons Media

Postby oln » 06 Mar 2011, 12:19

p0ss {l Wrote}:
It seems to me that up until this point the art in the OpenDungeons Project has been functioning in a fairly ad hoc way, I would suggest that there needs to be some strategic direction and a unified artistic vision, which can best be achieved by establishing content guidelines and todo/request lists. This appears to me to be a key priority if OpenDungeons is ever to move form being a techdemo into being an actual game.

In regards to content guidelines, I'd suggest having a look at the DungeonHack content creation guidelines as a starting point: http://dungeonhack.sourceforge.net/Art_Workflow_3D

I hope I haven't trod on too many toes, and if I have I hope it doesn't hurt too much because I feel much of this needs to be said. Goodluck and good day.


Thank you for a good block of constructive criticism. I agree with your post, guidelines are something we should prioritise for the next (that is 0.4.8) release.
I've looked at DungeonHack, and it is a good example of how things should be done. As they are also using Ogre, we should look at how they have implemented normal mapping so we can sort that out in OD as well. The mirrored UV maps is probably one of the reasons we haven't had that working yet, and is something we should try to fix.
User avatar
oln
 
Posts: 1020
Joined: 26 Oct 2010, 22:16
Location: Norway

Re: An Outsiders Perspective on the OpenDungeons Media

Postby Skorpio » 06 Mar 2011, 19:33

Yes, we're a bunch of foolish noobs, aren't we? Actually I'm already aware of the points you mentioned and I know there are a lot of things that need to be optimized. The poly/vertex counts are so low, because we had planned to have a very high number of models on the screen at the same time, that won't be very practical, though. We could raise the poly count to 2000, 2500 or even higher, if we go for a lower creature count. It's true that the poly count is less important with modern hardware, but RTS games still rely on models with a lower poly count. And not everybody has a high end pc.
The textures are so huge, because they are the source textures which haven't been scaled down yet. The materials cause the models to be shadeless atm, that's why everything looks so bright.

oln {l Wrote}:I've looked at DungeonHack, and it is a good example of how things should be done. As they are also using Ogre, we should look at how they have implemented normal mapping so we can sort that out in OD as well. The mirrored UV maps is probably one of the reasons we haven't had that working yet, and is something we should try to fix.

It's quite the contrary, the uv maps are still mirrored, because normal mapping isn't implemented yet. To be honest I'm quite frustrated, because I've been waiting for normal mapping for such a long time. Some creatures like the tentacle and troll have already unmirrored maps and could be used for testing. And unmirroring and baking the other textures shouldn't be a problem. BTW there are techniques to use mirrored normal maps, but they would also affect the performance negatively.
User avatar
Skorpio
OD Moderator
 
Posts: 775
Joined: 05 Dec 2009, 18:28

Re: An Outsiders Perspective on the OpenDungeons Media

Postby svenskmand » 06 Mar 2011, 19:52

Skorpio {l Wrote}:It's quite the contrary, the uv maps are still mirrored, because normal mapping isn't implemented yet. To be honest I'm quite frustrated, because I've been waiting for normal mapping for such a long time. Some creatures like the tentacle and troll have already unmirrored maps and could be used for testing. And unmirroring and baking the other textures shouldn't be a problem.

Yes we should prioritize normal maps in the next release. But I was under the impression that oln got them working recently?
Jamendo.com - The best music store on the net, uses CC licenses.
User avatar
svenskmand
OD Moderator
 
Posts: 1850
Joined: 09 Dec 2009, 00:07
Location: Denmark

Re: An Outsiders Perspective on the OpenDungeons Media

Postby oln » 06 Mar 2011, 19:55

Not properly, I could test with one of the two creatures with proper UV maps though, and maybe it would look right.
User avatar
oln
 
Posts: 1020
Joined: 26 Oct 2010, 22:16
Location: Norway

Re: An Outsiders Perspective on the OpenDungeons Media

Postby p0ss » 07 Mar 2011, 08:56

You are very wrong there, rendering performance is not proportional to the size of the textures being used.


Indeed, that section was hyperbole, it is not actually 100 times slower, however it is a massively inefficient use of resources.

I am glad you guys took this post so well, if you come up with a wishlist and todo list of art needed for this project I'd be glad to help out. If you have any specific resource requests it might be worth also posting them on OGA. As I said to oln in chat, I am one of the admins at OGA and we'd certainly be willing to help publicize your art bounty, however I think you probably need to come up with a clearer list of what needs doing first.
p0ss
 
Posts: 26
Joined: 18 Jan 2011, 02:10

Re: An Outsiders Perspective on the OpenDungeons Media

Postby StefanP.MUC » 07 Mar 2011, 17:42

I have to few knowledge about 3D art to comment on the technical aspects of the post, but I agree on the polygon number and variety of textures. Currently OpenDungeons looks not as good as DK2 did (IMHO) and DK2 is 12 years old. We shouldn't limit ourselves just because there are still some guys out there with low-end computers. By keeping the graphics down we win one new "low-end" player but lose ten "high-end" players that simply expect better graphics (because they know it from recent commercial games).
That doesn't mean we should have some specific "goal" about the graphics quality or even try to look better than the best commercial games, but the goal should be to look "as best as possible" and only "scale down" if, for example, an 4 year old mid-tech computer would have serious problems with the graphics.
StefanP.MUC
 

Re: An Outsiders Perspective on the OpenDungeons Media

Postby svenskmand » 07 Mar 2011, 18:16

blablub {l Wrote}:I have to few knowledge about 3D art to comment on the technical aspects of the post, but I agree on the polygon number and variety of textures. Currently OpenDungeons looks not as good as DK2 did (IMHO) and DK2 is 12 years old. We shouldn't limit ourselves just because there are still some guys out there with low-end computers. By keeping the graphics down we win one new "low-end" player but lose ten "high-end" players that simply expect better graphics (because they know it from recent commercial games).
That doesn't mean we should have some specific "goal" about the graphics quality or even try to look better than the best commercial games, but the goal should be to look "as best as possible" and only "scale down" if, for example, an 4 year old mid-tech computer would have serious problems with the graphics.

I very much agree on this. Why limit ourself and risk creating a inferior game?
Jamendo.com - The best music store on the net, uses CC licenses.
User avatar
svenskmand
OD Moderator
 
Posts: 1850
Joined: 09 Dec 2009, 00:07
Location: Denmark

Re: An Outsiders Perspective on the OpenDungeons Media

Postby TheAncientGoat » 07 Mar 2011, 18:48

Erm, I think Skorpio's models are really nice... He's done a lot of stuff for OD and we should support him, I don't think there's anything really wrong with his stuff. The terrain is too low poly though, as well as the mining pick..
User avatar
TheAncientGoat
Community Moderator
 
Posts: 518
Joined: 27 Dec 2009, 19:06

Re: An Outsiders Perspective on the OpenDungeons Media

Postby svenskmand » 07 Mar 2011, 19:36

Yes Skorpio is mainly the reason why we have such greate models at all :) He is very skilled indeed :)
Jamendo.com - The best music store on the net, uses CC licenses.
User avatar
svenskmand
OD Moderator
 
Posts: 1850
Joined: 09 Dec 2009, 00:07
Location: Denmark

Re: An Outsiders Perspective on the OpenDungeons Media

Postby StefanP.MUC » 07 Mar 2011, 19:45

The models are great, no question. But the original models are stripped down before getting into the game. We shouldn't cut away to much. It should still look good if the player zooms into the dungeon.
While it's true that strategy games have many models at the same time (the argument for low-poly I hear most in the FOSS sector), it's still so that only a small amount of them is visible at the same time (and only the visible ones get rendered).

The way should NOT be: "strip down the graphics until everyone can play".
It should be like: "try creating the best graphics until a notable bottleneck for the majority of users is reached, then try to optimize the performance of the engine and graphics, and only as the very last step, if nothing else works, strip down the graphics"

I have an almost 4 years old notebook and installed a VirtualBox VM with Ubuntu (strongly cut down memory and not very good drivers, this way it's probably like a 12+ years old computer) and OpenDungeons still runs pretty well in it (it's hard on the "stuttering edge", but it works).
And if the optimizations from p0ss really brings more performance, then I see absolutely no reason, why we shouldn't raise our "graphical wishes" some way up.
StefanP.MUC
 

Re: An Outsiders Perspective on the OpenDungeons Media

Postby svenskmand » 07 Mar 2011, 19:54

blablub {l Wrote}:The models are great, no question. But the original models are stripped down before getting into the game. We shouldn't cut away to much. It should still look good if the player zooms into the dungeon.
While it's true that strategy games have many models at the same time (the argument for low-poly I hear most in the FOSS sector), it's still so that only a small amount of them is visible at the same time (and only the visible ones get rendered).

The way should NOT be: "strip down the graphics until everyone can play".
It should be like: "try creating the best graphics until a notable bottleneck for the majority of users is reached, then try to optimize the performance of the engine and graphics, and only as the very last step, if nothing else works, strip down the graphics"

I have an almost 4 years old notebook and installed a VirtualBox VM with Ubuntu (strongly cut down memory and not very good drivers, this way it's probably like a 12+ years old computer) and OpenDungeons still runs pretty well in it (it's hard on the "stuttering edge", but it works).
And if the optimizations from p0ss really brings more performance, then I see absolutely no reason, why we shouldn't raise our "graphical wishes" some way up.

I agree.
Jamendo.com - The best music store on the net, uses CC licenses.
User avatar
svenskmand
OD Moderator
 
Posts: 1850
Joined: 09 Dec 2009, 00:07
Location: Denmark

Re: An Outsiders Perspective on the OpenDungeons Media

Postby andrewbuck » 07 Mar 2011, 21:27

One thing that Ogre supports that we could look into as well is Level of Detail scaling. Ogre supports LOD for both meshes, and textures. So small, far away meshes can use low polygon meshes and also smaller texture files/and or skipping things like normal mapping, etc. The Ogre engine does this basically automatically, you just have to tell it what the various options it has to use and which ones it should use for "important" models. For materials, you can specify multiple technique sections in a material script, one with normal mapping and one without. Then ogre will automatically use the better shading technique when the mesh in question is closer to the camera.

I think p0ss is correct that we shouldn't over constrain ourselves and don't try to be too skimpy on polycounts, etc, but I also think that in the end our focus on this will help the performance of the game. Having lower polygon meshes available (even if higher poly ones are built later) allows us to make better use of the LOD functions and just speeds up the game in general. The current pipeline of making a low poly mesh, and then using that as a basis for a high poly model made with a sculpting tool seems to be working pretty well and has the advantage that in the end we end up with both a highly optimized, low-poly mesh, and a high poly one. It also allows us to see/play with/comment on/redesign the low-poly mesh before we put too much time into high resolution models and textures before finding out we want to change something.

-Buck
andrewbuck
OD Moderator
 
Posts: 563
Joined: 20 Dec 2009, 01:42

Re: An Outsiders Perspective on the OpenDungeons Media

Postby charlie » 07 Mar 2011, 23:08

I would suggest that the current "low poly" nature of OpenDungeons stems from the youthfulness of the project and inexperience of the contributors. As the project grows and the boundaries of development are explored, implementing LOD support and iteratively improving the existing models is a natural consequence. It's more a decision of what level do you want to commit to, and when to commit to it - bearing in mind the restricted resources and the current incomplete state of the content. Better to put in a bunch of low poly but complete content first then attack higher detail content later, no? Or just leave it up to the artists as to just how detailed they wish to go.
Free Gamer - it's the dogz
Vexi - web UI platform
User avatar
charlie
Global Moderator
 
Posts: 2130
Joined: 02 Dec 2009, 11:56
Location: Manchester, UK

Who is online

Users browsing this forum: No registered users and 1 guest