Building a carriage (and a lot, lot, lot of questions).

Discussion of rolling-stock creation & re-painting.

Building a carriage (and a lot, lot, lot of questions).

Unread postby TrabantDeLuxe » Thu Jun 11, 2015 1:07 pm

Hi,

I'm in the process of creating a series of compartment coaches built for the Dutch State Railways (Staats-Spoorwegen). I'm rather amazed at how they've turned out, but I feel that I kinda need some help on resource management. I'm getting this feeling that I'm going way over budget, texture wise, and polygon wise. Below is a screenshot of one of these coaches:

Image

Type B6c compartment coach, built around 1890.


I'm working on a few more coaches of the same appearance. I'd really appreciate some input on 'best practices'. The polygon count is in the 10000 - 12000 range, and the triangle count is about twice these values. How would you manage LOD's? Is there a rule of thumb for poly counts at say 100 meters, 500 meters, and 2000 meters?

Textures are another headache. Almost everything is textured using the TrainLightBumpSpecMask.fx shader, giving diffuse, normal, specular, and ambient occlusion. The AO uses a seperate unwrapping, because many parts are overlapped on the diffuse. The normal and specular are edited from the diffuse.

  • Body: TrainLightBumpSpecMask.fx. Using 2048 by 2048 diffuse, and a 2048 by 2048 normal, with specular alpha. This material is unique for each coach.
  • Frame: TrainLightBumpSpecMask.fx. Using 2048 by 2048 diffuse, and a 2048 by 2048 normal, with specular alpha. This material is universal, and can even be adapted for twin-axled coaches.
  • Windows: Trainglass.fx. 256 by 256 diffuse, with translucent alpha. Also a 32 by 32 dummy texture. This texture can be shared amongst coaches.
  • Interior: TexDiff. I've got one 1024 by 1024 that has bits of 1st, 2nd and 3rd class interior, so this is a common texture.
  • Decals (not shown): Decal are created by two 512 by 512 textures. These may be shared amongst passenger carriages, luggage vans, freight, what you like. Everything with letters on 'em.

Would any of the experts be so kind as to comment on these statistics? Another thing that I'm not entirely sure of, is how Train Simulator handles common textures. Say I'd have a train of 5 different carriages, all variations of the above. Would the 'common' textures be loaded just once, or would I still get a draw call for every instance of the texture in the game? The latter of course rather undermines my enterprise to keep textures common...

And yeah, LODs. Is it good practice to just ignore the AO unwrap as I cut and weld to reduce polycount? That's what I've been doing now. I'd love some screenshots from 3ds Max or so showing a distant LOD!

Cheers,
-deLuxe.
TrabantDeLuxe
 

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby mrennie » Thu Jun 11, 2015 2:15 pm

Here's one tip.

Put the smaller details, all of a similar size, such as nuts and bolts, into their own groups (or, if you need to, separate child objects - that can hep if the model is getting too big to handle in 3DCrafter, which I use) with a unique LOD value. Set that LOD to a value such as 0200. Place the model in the game. In the 8 view, move backwards. Do you see those details suddenly vanish, or do they disappear simply because it looks like you're too far away to see them? If it's the former, the LOD value is too low, so increase it. If it's the latter, the LOD value might be too high, so decrease it. Repeat the process until you hit the optimum value for the LOD for that particular group of details. The optimum is where you don't notice the details suddenly disappear even though in reality they have.

It works best if you put details of a similar size in the same group (so that they all have the same LOD distance value). That's how I did the details on the FEF-3 (for example, all the rivets on the tender are in the same group). The reason it works so well is that when you get closer to the loco, the amount of background stuff in the entire scene that you can see reduces, so there's more video budget to show the close up details without reducing the fps. As you move further away, more stuff in the scene surrounding the loco comes into view, but you maintain the fps because all those details on the loco that would be too small to see anyway simply don't get drawn, thanks to the LOD.
User avatar
mrennie
 
Posts: 3214
Joined: Wed May 30, 2012 12:22 pm

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby TrabantDeLuxe » Fri Jun 12, 2015 2:00 am

Thanks. I just did some experimentation with that, and an interesting observation I might share. It seems that the game engine automagically adjusts the LOD level based on distance (as it's supposed!), as well as for the field of fiew. When you scroll up for that telephoto look, you get a higher LOD! Pretty neat.
TrabantDeLuxe
 

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby LordMannu » Fri Jun 12, 2015 6:36 am

Your Poly count still into the MSTS tolerance range. You can easly have twice of that amount and still having a great model. My engines are in the range of 80000 polys.
Regarding to LODs, keep the details in its own group or node is a good idea. I keep the small details until 75 or 100 meters, and the furthest until 1500 meters.
And for the textures, all looks ok to me. You are doing a great job.
User avatar
LordMannu
 
Posts: 334
Joined: Tue Feb 17, 2009 8:51 am
Location: Buenos Aires, Argentina

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby TrabantDeLuxe » Fri Jun 12, 2015 7:04 am

So, basically I'd be right to ignore the developer docs on that point? I like detail, but not at the expense of a slideshow fest. 'Test it yourself' would be a bit difficult - I'm doing all my modding on a 3 year old laptop.
TrabantDeLuxe
 

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby mrennie » Fri Jun 12, 2015 8:12 am

TrabantDeLuxe wrote:So, basically I'd be right to ignore the developer docs on that point? I like detail, but not at the expense of a slideshow fest. 'Test it yourself' would be a bit difficult - I'm doing all my modding on a 3 year old laptop.


Absolutely! It would be a massive mistake to follow the budget guidelines in the developer docs. Those were written in the days of the Pentium 4 and <1GB RAM, or worse!! I actually still find it sad that some developers still think they're obliged to adhere to those nonsensical limits ... their models suffer as a result. It took models from the likes of Ricardo, Mannu, VRC, myself and others to show that "more is more". What I personally dislike is when I see something on a model that looks ok from a distance, and then you go in close to look at it in detail, and discover that it isn't even modelled in 3D, or that it's not even properly rounded. I've deleted models from my hard drive for that very reason. It often happens with large "details" too, such as driving wheels!! There's nothing worse than seeing driving wheels that look like something out of The Flintstones.

Testing on a 3 year old laptop is an excellent idea. I did a lot of the modelling of the FEF-3 on an even older laptop, albeit on lower settings than on my main PC (which itself is 6 years old), to make sure that it would be ok in average machines. In fact, what I found was that I really needn't have spent so much time on the "SD" ("Standard Definition", which I had originally called "Reduced Detail"), because even my "High Definition" model works fine on the old laptop.

When I have time, I'll examine the FEF-3 polycount. It's a bit difficult because it's made of so many child objects, but it's pretty big anyway. What I found was that nowadays, with even average graphics cards (when I was modelling the Connie and FEF-3, I was still using a single GTX295!!!), the polycount doesn't have much of an impact. What makes more of a difference is the size and number of texture files, but even there, things have advanced enormously.
User avatar
mrennie
 
Posts: 3214
Joined: Wed May 30, 2012 12:22 pm

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby bnsfsubdivision » Fri Jun 12, 2015 10:12 am

mrennie wrote:
TrabantDeLuxe wrote:So, basically I'd be right to ignore the developer docs on that point? I like detail, but not at the expense of a slideshow fest. 'Test it yourself' would be a bit difficult - I'm doing all my modding on a 3 year old laptop.


Absolutely! It would be a massive mistake to follow the budget guidelines in the developer docs. Those were written in the days of the Pentium 4 and <1GB RAM, or worse!! I actually still find it sad that some developers still think they're obliged to adhere to those nonsensical limits ... their models suffer as a result. It took models from the likes of Ricardo, Mannu, VRC, myself and others to show that "more is more". What I personally dislike is when I see something on a model that looks ok from a distance, and then you go in close to look at it in detail, and discover that it isn't even modelled in 3D, or that it's not even properly rounded. I've deleted models from my hard drive for that very reason. It often happens with large "details" too, such as driving wheels!! There's nothing worse than seeing driving wheels that look like something out of The Flintstones.

Testing on a 3 year old laptop is an excellent idea. I did a lot of the modelling of the FEF-3 on an even older laptop, albeit on lower settings than on my main PC (which itself is 6 years old), to make sure that it would be ok in average machines. In fact, what I found was that I really needn't have spent so much time on the "SD" ("Standard Definition", which I had originally called "Reduced Detail"), because even my "High Definition" model works fine on the old laptop.

When I have time, I'll examine the FEF-3 polycount. It's a bit difficult because it's made of so many child objects, but it's pretty big anyway. What I found was that nowadays, with even average graphics cards (when I was modelling the Connie and FEF-3, I was still using a single GTX295!!!), the polycount doesn't have much of an impact. What makes more of a difference is the size and number of texture files, but even there, things have advanced enormously.

!!*ok*!!
Kevin Schulz
Artist - Searchlight Simulations
User avatar
bnsfsubdivision
 
Posts: 822
Joined: Sun Feb 20, 2011 5:27 am

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby TrabantDeLuxe » Fri Jun 12, 2015 11:00 am

bnsfsubdivision wrote:
mrennie wrote:
TrabantDeLuxe wrote:So, basically I'd be right to ignore the developer docs on that point? I like detail, but not at the expense of a slideshow fest. 'Test it yourself' would be a bit difficult - I'm doing all my modding on a 3 year old laptop.


Absolutely! It would be a massive mistake to follow the budget guidelines in the developer docs. Those were written in the days of the Pentium 4 and <1GB RAM, or worse!! I actually still find it sad that some developers still think they're obliged to adhere to those nonsensical limits ... their models suffer as a result. It took models from the likes of Ricardo, Mannu, VRC, myself and others to show that "more is more". What I personally dislike is when I see something on a model that looks ok from a distance, and then you go in close to look at it in detail, and discover that it isn't even modelled in 3D, or that it's not even properly rounded. I've deleted models from my hard drive for that very reason. It often happens with large "details" too, such as driving wheels!! There's nothing worse than seeing driving wheels that look like something out of The Flintstones.

Testing on a 3 year old laptop is an excellent idea. I did a lot of the modelling of the FEF-3 on an even older laptop, albeit on lower settings than on my main PC (which itself is 6 years old), to make sure that it would be ok in average machines. In fact, what I found was that I really needn't have spent so much time on the "SD" ("Standard Definition", which I had originally called "Reduced Detail"), because even my "High Definition" model works fine on the old laptop.

When I have time, I'll examine the FEF-3 polycount. It's a bit difficult because it's made of so many child objects, but it's pretty big anyway. What I found was that nowadays, with even average graphics cards (when I was modelling the Connie and FEF-3, I was still using a single GTX295!!!), the polycount doesn't have much of an impact. What makes more of a difference is the size and number of texture files, but even there, things have advanced enormously.

!!*ok*!!


!!*ok*!! I'll second that. Also, I think it's kinda awesome that a developer of commercial add-ons shares his knowledge. You sir, are a gentleman and a scholar.
TrabantDeLuxe
 

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby mrennie » Fri Jun 12, 2015 11:16 am

TrabantDeLuxe wrote: Also, I think it's kinda awesome that a developer of commercial add-ons shares his knowledge. You sir, are a gentleman and a scholar.


*!!thnx!!*
User avatar
mrennie
 
Posts: 3214
Joined: Wed May 30, 2012 12:22 pm

Re: Building a carriage (and a lot, lot, lot of questions).

Unread postby TrabantDeLuxe » Sun Jun 14, 2015 3:11 pm

I've just done the LODS, and I thought I'd show them for those who are interested.

Image

What may be seen is, that just as Mike suggested, all the detailing goes at 100 meters. I've concluded that after 250 meters, the wheels themselves are barely visible, so they are removed. The interior is removed at 150 meters.

Next question: What UV argument in the TrainLightBumpSpecMask.fx controls what? At this moment I'm just fiddling about, and it would really speed up the workflow actually knowing what I'm tweaking.
TrabantDeLuxe
 


Return to Rolling-Stock Design

Who is online

Users browsing this forum: No registered users and 1 guest