So, what are you working on right now?

[Springheel]

The handle seems to be missing.

 

That's mainly because it's frob-lit. You lose specularity and most of the normalmap when the froblight is on.

[Diego]

I could have, but I'm still trying to keep a budget..

I think it's fine that way. It's as high poly as is suitable for a game engine

(the shot of iron stuff above is editor due to game crash on screenshot but they do have spec in game, but it's a white spec which makes it too shiny imo)

I'm not sure if I understand what you mean.

2: I removed spec off a wood on the old chest just the other day. Basically any wood in game that isn't highly polished looks like plastic with spec.

A lot of things shine, but definitely not all. Unless wood is polished it doesn't have spec, even when wet, if it has any grain at all it is a matte material.

I don't think I ever used a material with zero specular. And I work with CG for more than a decade. What's happening here is that the engine doesn't allow control over this specular highlight, so everything has the same concentration only with different intensities. With this limitation you're absolutely right, it's probably better to remove specular from some materials, if they look plastic at this fixed concentration.

[Diego]

That's mainly because it's frob-lit. You lose specularity and most of the normalmap when the froblight is on.

Nah.. I'm only frobbing one. The other also doesn't have a handle. I know because I was there when I took the shot

[Springheel]

I'm missing something then, because all I can see are ones with handles, one that's froblit, and one that's too dark to see anything at all.

[Diego]

It's not that dark.. is it?

[Tels]

I could have, but I'm still trying to keep a budget...

 

For example, the 'fancy' single rail.. two post with 3 curls is 402 polys alone with a shadow of 214. two post with circle is 304, shadow 176. So if an author used that one 12 stairs... Well you know math.

 

The sign hanger is 814 polys, with a shadow of 372.

 

I'm trying to do higher polys stuff, but still trying to stay on budget. To make them look much rounder I'd have to bump those counts up quite a bit. And redo a lot of work, I just don't see it being necessary personally.

 

Well, if I hadn't seen it nec. I would have asked

 

As for the polycount, I am not exactly seeing how we objectively "judge" it. All we are using all over the place are one-size-fits-all numbers (where everybody seems to agree that xxx polies is "too many" and yyy plies is "not enough", but there are rarely any scientific measurements. D3 can push a lot of polies, so if a rail uses 5000 for a closeup, but 50 in the distance might be much better if it uses 500 all the time.

 

Now, I am not saying that you are right or wrong with your numbers, but I am just saying that when I can see the polygon edges even from that far away, it has not enough for a rounded shape

 

So, anyway, one idea I had: would it be possible to use the trick you (I think it was you) used with some gears, where you have basically two flat sides with a texture and then only the "slim edges" are polygons? Or would that not save much and create other problems?

 

Anyway, this is one of the "everyone has their favourite opinion" topics, and nothing what I say here is going to change yours, anway. So we might as well skip that discussion....

 

I was going to ask though if it's possible to use a skin as the LOD stage instead of a model. The shadow has got the same side profile, so the LOD stage could just be a skin swap where the iron goes to the shadow mesh and nodraw is placed on the iron. And the shadow just goes away.

 

I don't think I have implemented it. But you could simply create a second model. The problem with skin changes is, that when they change which material cast shadows, the engine does a very expensive recomputation of the shadow edges, anyway. So swapping the model might actualy be a faster way to switch - and then you also don't have the problem that "nodraw" faces are still active for collision detection (e.g a skin change just makes them invisible - but it does not remove them from the model as far a the engine is concerned).

[Baddcog]

Alright, I'll just export the shadow mesh as a model.

 

I can't used the alpha plane trick on these as I used a diamond profile instead of square. And I think we probably lose as much performance with the alpha as it would take in polys (gears might be a good trade because there are so many teeth. honestly I don't know which is better in the long run - alpha or more polys).

 

The numbers are pretty arbitrary, it's just one of those things were as a modeller you have to budget and try to get the most bang for the buck. Sure we all want things to look super awesome but gotta try and not get carried way too (I recall someone giving me crap for doing a very high poly syringe even though it had a lowpoly shadow).

For TDM we've been trying to hit certain numbers that were already pretty far above Doom3 specs and on my later models I've even been trying to push that a bit using a lowpo shadow. Still trying to keep from getting out of control with it.

 

LOD certainly weighs in on that consideration too. One concern I still have with this is the system we have for models/entities/prefabs. It's easy enough for authors to learn there is an entity system, but most authors still probably don't look for model entites for this purpose and honestly it's been awhile since I've been doing anything and I never set them up (at work atm so I can't look).

From what I recall an author needs to create an entity to have LOD correct?

 

So I'll round out some edges, save that as railing_LOD_hi and I'll put them into a fencing/LOD folder with the lows. So if an author just uses the model they will use medium detail. If they use LOD entity they will get full stages. That's the best compromise imo.

[Tels]

Alright, I'll just export the shadow mesh as a model.

 

I can't used the alpha plane trick on these as I used a diamond profile instead of square.

 

Ah ok.

 

And I think we probably lose as much performance with the alpha as it would take in polys (gears might be a good trade because there are so many teeth. honestly I don't know which is better in the long run - alpha or more polys).

 

Yeah, it boils down to a number of factors. Rendering one polygon with alpha test means only one polygon (which is good for the CPU as D3 is CPU bound), but it means a few more pixels in the pixel shader - which is bad for the GPU.

 

I *think* for a CPU bound engine as D3 alpha maps should be better than polygons - but I have no numbers to back that up.

 

All I can say is that I have had scenes with 2 million polygons, and while the FPS was low, it didn't blow up

 

The numbers are pretty arbitrary, it's just one of those things were as a modeller you have to budget and try to get the most bang for the buck. Sure we all want things to look super awesome but gotta try and not get carried way too (I recall someone giving me crap for doing a very high poly syringe even though it had a lowpoly shadow).

 

Truth to be told, with LOD I'd like the high-poly for real closeups and the low-poly for normal views, but with a one-size-fits-all, I'd also say a bit lower is better (not because I amworried about performance, but because it will just stand out from everything else that is just a 7-sided cylinder

 

For TDM we've been trying to hit certain numbers that were already pretty far above Doom3 specs and on my later models I've even been trying to push that a bit using a lowpo shadow. Still trying to keep from getting out of control with it.

 

LOD certainly weighs in on that consideration too. One concern I still have with this is the system we have for models/entities/prefabs. It's easy enough for authors to learn there is an entity system, but most authors still probably don't look for model entites for this purpose and honestly it's been awhile since I've been doing anything and I never set them up (at work atm so I can't look).

From what I recall an author needs to create an entity to have LOD correct?

 

Yes. The engine itself cannot change a model w/o an entity to tell it so (I guess now with the source code, we could build this directly into the renderer. Tho we don't have no-one who works on this stuff at all)

 

So I'll round out some edges, save that as railing_LOD_hi and I'll put them into a fencing/LOD folder with the lows. So if an author just uses the model they will use medium detail. If they use LOD entity they will get full stages. That's the best compromise imo.

 

Yes, that sounds like a very good plan.

[Nosslak]

1: The spec map almost has to be black to have a 'sane' amount of specular.

So I've done some work that hasn't been implemented mod wide yet on metals, basically adding their color (pink for copper, yellow for gold) to spec maps. For it to look good the yellow is almost dark brown.

It takes a lot of time to tweak it just right. I noticed yesterday as I was playing and saw a gold skull and it looked polished to the point of loot glow (must've been where I left off as I just dug the files out again), and we don't want that.

For my recent pimpings of the spyglass, alembic and the mask textures as well I did it the other way around. I made a pretty dark diffuse and then a very bright specular with a lot of contrast for the specmap. I had read that that was a good idea (on Polycount) so I tried it and the results looked great. I did render them in Marmoset though, where I can also change the strength and intensity of the spec so I might need to adjust them. I'll try to test them in-game either today or tomorrow.

 

4:We've been through the ambient bake thing quite a few times. A lot of stuff was made before anyone knew what AO was. It was just starting to be used around the time the mod started. So that would mean going back and doing tons of work. Then I started using existing shaders as much as possible on as many objects as possible to save space and tex loading in game. Most of this stuff wasn't UV'ed for AO, so tons of that would need done and baking.making a lot of object specific textures...bleh

What's going on with SSAO? Last time I heard anything about that we needed the source to do that without such a big performance hit (if I recall correctly you needed to render a new depth-buffer, which is expensive). That would solve at least some of these problems without redoing any models.

 

I don't think anyone answered my question: Is it possible to use glossmaps or a single glossvalue somewhat easily?

[Baddcog]

I've never heard anything about a gloss map, but Serp was saying he might be able to change the spec intensity which would be a great improvement. Polished metals need a sharp falloff, things like silk need a broad falloff. Of course that's best for a material by material setting. An object that has different materials on one texture (wood barrel metal rings) wouldn't work as well. (but since wood has very low spec then the sharper highlight on a per object basis might not even show on the wood, but the metal would glint better.

 

SSAO would be a great improvement visually, but we need coders...I imagine it's a big job too. Our best bet is probably if someone makes a good solution for Doom3 that we are able to fold into TDM.

 

Not sure how the darker diffuse/spec thing would look in game. Would like to see it, but I imagine in a game that is fairly dark to begin with having darker diffuse will just make stuff harder to see. And Doom3 looks plasticy because of overdone specular maps. So I'm sceptical but will wait to make final judgement

I have a feeling that marmoset rendering is probably better than Dooom3's though.

[Tels]

What's going on with SSAO? Last time I heard anything about that we needed the source to do that without such a big performance hit (if I recall correctly you needed to render a new depth-buffer, which is expensive). That would solve at least some of these problems without redoing any models.

 

We don't have anyone who could (or does) work on this, even if somebody else implemented it in D3, we'd need someone to merge their work in TDM. I wouldn't bet on it, even tho these things would all be very nice.

[Nosslak]

I've never heard anything about a gloss map, but Serp was saying he might be able to change the spec intensity which would be a great improvement. Polished metals need a sharp falloff, things like silk need a broad falloff. Of course that's best for a material by material setting. An object that has different materials on one texture (wood barrel metal rings) wouldn't work as well. (but since wood has very low spec then the sharper highlight on a per object basis might not even show on the wood, but the metal would glint better.

AFAIK the gloss map is often stored in the alpha channel as a b/w map. I haven't used a lot of engines (Marmoset and just barely Doom 3 and UDK) and I know that this is how Marmoset does it and I think I've read that this is how UDK or Cryengine does it too (I might be completely wrong but I think I read about this on Polycount). Blender has support for it as well (don't know if the Blender Game Engine supports it as it is supposedly kinda crap).

 

Not sure how the darker diffuse/spec thing would look in game. Would like to see it, but I imagine in a game that is fairly dark to begin with having darker diffuse will just make stuff harder to see. And Doom3 looks plasticy because of overdone specular maps. So I'm sceptical but will wait to make final judgement

I have a feeling that marmoset rendering is probably better than Dooom3's though.

Yeah, I have no doubt that Marmoset renders better than Doom 3, but it is very quick and easy to set-up and yield great results so that's why I use it.

 

We don't have anyone who could (or does) work on this, even if somebody else implemented it in D3, we'd need someone to merge their work in TDM. I wouldn't bet on it, even tho these things would all be very nice.

I don't know much about what kind of work any of you programmers do, but isn't this kind of what JC_Denton does?

[Tels]

I don't know much about what kind of work any of you programmers do, but isn't this kind of what JC_Denton does?

 

I was about to post that he hasn't visited the forums for a year or so, just to see he was here 6 days ago:

 

http://forums.thedarkmod.com/user/3576-j-c-denton/

 

So, JC Denton, if you read this, is this something that interests you?

 

I would still not get my hopes up, as JC hasn't committed or being active for a long time in TDM. As for the others, there are essentially only 4 programmers left (greebo, grayman, serpentine and me), and grayman is about the only one who does most of the work but he is more or less AI only. greebo does some little work from time to time but he is more focused on DR most of the time. Serpentine is the one whos coming closet to the renderer, but I am not sure if he has the time an energy for that sort of work. I am useless in this regard.

[Baddcog]

added a couple skillet models and a soup ladel.

 

Added a high and low lod for my new wrought iron sign hanger. defs for all. Though I still don't have doom 3 installed so I haven't texted the collision models or lod in game. They should be fine though.

[Serpentine]

I don't quite have any pretty pictures, however I've been working on the render back-end; Tho I still haven't got my head around it completely, I've found a fixed a good few areas where there was expensive branching going. I managed to squeeze 15fps out of the worst spot on stablans last night, and at the moment I'm just moving around looking for more. I don't quite understand how it all glues together, and since I am both bad with code and completely new to engine internals, it'll be a while 'til anything nifty happens.

 

So - working on performance so that we can get sparkly stuff later on, rather than adding them early and then running into soul crushing performance overheads. Oh, and if anyone is bored and wants to start rewriting our shaders into glsl, feel free to get in touch.

[Melan]

That's very important work - maybe less tangible and instant-gratification than a map of a model, but if TDM could improve performance by just three or four small steps, that would open it up for a lot of people with weaker configs.

[Tels]

Someday I hope we can also rewrite the "combine models" system used by SEED to just use instancing on the graphics card. That would remove a lot of overhead upon start of the mission and when moving (e.g. when the combined models are build) and also render a bit faster and use a lot less memory.

[nbohr1more]

Not to pressure anyone but GameDevGoro has built an engine (more or less) according to his blog. Perhaps he'd be willing to have a couple of PM sessions with Serpentine (et al.)?

 

As for SSAO, has anyone spoken to Rebb about this. It seems that he has the aptitude?

[RPGista]

Baddcog and Serpentine, some awesome work you guys are doing.

 

I was working on a bunch of models for my mission, some simple stone benches to be precise, and it was looking ok in the editor, but when I tried to export them I was getting this strange messed up surfaces that wont render correctly (tried it on Rhino and Kerkythea), its probably something really obvious, but I couldnt find the time to be looking for answers and kinda of gave up on the whole attempt - might aswel ask one of you experienced modelers, whats up with these models? (The last pic is how it looks on other editors after exporting from Hexagon)

 

 

 

Edited February 26, 2012 by RPGista

[Baddcog]

I don't know Hexagon or Kerky..

So not sure what rendering issues they may have.

 

Also not sure what you are getting in DR, does it look like above? that render definitely uses vert lighting so more verts will make it look better, but probably not needed in TDM.

 

Doom3 doesn't do smoothing groups, a bit of a pain because you gotta model then split edges to get sharp edges. So for the chips I'd split all the edges so they'll be sharp. Then the bevels you have should keep the lighting looking decent around the smoothed corners. (give your whole model one smooth group).

Also, in DR and the model viewer, they won't look all that hot, they do look better in game.

[RPGista]

Yeah, sorry, Hexagon is the 3d program I was using to create the benches, and kerkythea is just an open renderer that I sometimes use for other things. The problem is that I was getting messed up surfaces when I was trying to render outside Hexagon (just to show them here, actually), I dont know if the models themselves were messed up on the creation process, if the exporting is buggy or if there are settings I didnt set up correctly (and got duplicated surfaces or something). I havent even tried putting them into the game yet, Im just starting at this, really. But anyway, thanks for the tips, will keep them in mind.

[Baddcog]

Well, wrapping it up today I think.

 

Got some new items for the mad scientists. holders for the flasks (which were modified/improved). And a candle lit bunsen burner entity to heat up their contents (I need to make some vials with liquids [probably using the player potion materials]). A distiller to distill liquids (yeah, the flask is similar but...)

 

 

Enlarged some smoke stacks that just always seem small in maps when I play (though I thought they were proper size when I made them- I'm finding a few items of mine that seem small when I see them in actual maps )

 

Changed the handles on my trophy loot since they were always dreadful. Added a few polys, but the shadow is about the same, and I optimized it some so it's close to even. Added an LOD for it too at 300 units, which should get a lot of use in tomb areas.

 

And last but not least, a little something for the ladies, a watering can to tend their flowers with.

 

(most of these tools share the various skins seen on the bunsen burner.)

[Bikerdude]

Those alternate smoke stakes will come in handy :-) and that watering can makes me think we need more gardening tools...

[Melan]

The watering can looks meh. The large smokestacks, though, are a godsend; the standard version is positively tiny.

[rich_is_bored]

@RPGista:

 

Your mesh is comprised largely of n-sided polygons and quads.

 

Game engines don't like polygons with more than three sides. They will split n-sided polygons and quads into triangles. You should do this in your modeling package instead so you can specify where the new edges end up. Your illumination problems stem from allowing other programs to decide how to triangulate your mesh.

 

Game engines also render polygon objects using smooth shading. The first two screenshots you've posted the mesh is flat shaded. By applying smooth shading to the object in your modeling application you are afforded an opportunity to spot and correct problem areas before you export the object.

Edited February 27, 2012 by rich_is_bored