Hi Guys,
Im after a tutorial for baking textures on an exterior architecural scene.
Does anyone have one or know of one that is around??
Thanks in advance!
hey jacks,
know this is an old thread but was wondering if u did find any tuts of texture baking??
Yep, this would be good to have. Anyone have any ideas? Always wondered about texture baking.
One of the chapters on my new DVD is about texture baking… but to be honest, I only see limited reasons why you would want to use it.
dansimon- i never did find any tutorials
booh, yeah. bakeing would be useful for larger scense where u have large surrounding geometry like misc houseing and terrain. guess it would speed up renders a bit.
:lol: for game level design, web3d and for all real time use… is that enough? 
Hi Guys,
If you’re still interested, here’s some useful tips I’ve found whilst baking
textures:
* Only bother baking the largest and most obvious surfaces, such as
floors, walls, ceilings. Don’t bother trying to bake the whole scene. It
will take a month.
* It’s always good to keep your baked textures sq. ie. 256x256, 512x512,
1024x1024. I think it’s a video card display issue. Someone correct me
if I’m wrong. So subdivide your mesh into large square surfaces.
* Using subdivided square meshes also means your maximising the area
of the baked texture in the saved image. The saved image is always
square. If you baked a rectangle for example, you’d be wasting half the
area in the baked texture with black.
* Definitely use Max’s baking engine.
* If you’ve baked an object, don’t like the result, delete the automatic
uvw unwrap from the modifier before re-baking. I’ve noticed it can
cause baking errors for future calculations.
Predominantly we use baking for Real Time simulation software, but have
started using it for very large architecturals. Particularly interior scenes
where there’s a lot of lighting required. Takes time to setup, but saves a
lot in render times.
SunnyC
ivolve studios
* Only bother baking the largest and most obvious surfaces, such as
floors, walls, ceilings. Don’t bother trying to bake the whole scene. It
will take a month.
That’s sure that backing all scene objects require more work particulary a very precise UVMAP for the backing, but that’s the price you have to pay for a very realistic result.
* It’s always good to keep your baked textures sq. ie. 256x256, 512x512,
1024x1024. I think it’s a video card display issue. Someone correct me
if I’m wrong. So subdivide your mesh into large square surfaces.
Yes but today’s video card support at least 2048x2048 and a lot of them 4096x4096 and one texture of 2048x2048 is allways a better choice than 16 texture of 512x512.
* Using subdivided square meshes also means your maximising the area
of the baked texture in the saved image. The saved image is always
square. If you baked a rectangle for example, you’d be wasting half the
area in the baked texture with black.
No. Subdivising your geometry like you say would introduce a lot of seams in the backed texture and so more potential texture leaks.
With a good packing you should not lost a large amount of space in the backed texture. But again packing uv cant be 100% automated if you want very good efficiency.
here is the famous SIbenik cathedrale 100% backed for realtime 3D walkthrough (near 50 000 polys after optimization) only one lightmap of 2048x2048 cover the entire geometry (plus of course texture for base material) :
(this is still a Work In Progress )
and the lightmap (here reduced to 1024x1024) :
what do you mean with: “only one lightmap of 2048x2048 cover the entire geometry (plus of course texture for base material)”
what’s the setup for your material?
do I have to put the *.tga to diffuse slot with 100% self-illumination?
what kind of map do I have to render? complete, diffuse, or…?
best regards
themaxxer
realtime 3d used to require power-of-2-sized textures
this has changed but they are still much faster
maximum texture size for one texture depends on your graphics card and driver
geforces with current drivers support 4096x4096
btw: power of 2 means both sides have to be a power of two but not necessarily the same (so no need for square textures)
yes, one larger texture for all objects is faster than one smaller texture for each of them
but you will get into trouble with spacing of the different areas (a lot of wasted texture space that increases with the number of different non-continuous areas in the texture)
so it might be worthwhile to use more than one texture but have those laid out well
there are several limiting factors that determine the speed of realtime 3d
fill rate, transfer rate, cpu time, memory bandwidth, pixel shaders, vertex shaders, rasterization.
textures are only a part of the problem
example: physique (character studio) animation of a high-density mesh in max
cpu-usage 100% (mesh topology changes and has to be sent to graphics card every frame), no matter how good you bake the textures, it will alway be limited by the cpu
if you are interested you might want to read this stuff:
http://developer.nvidia.com/docs/IO/8343/Performance-Optimisation.pdf
http://www.ati.com/developer/dx9/ATI-DX9\_Optimization.pdf
basicly, it all depends on your realtime engine and target audience (which card and cpu do they use?)
if you have access to the code (self-written or open source) you have more influence (or your programmer has
ask your developers (coders) for advice what they think works best, it’s a complex matter and they (should) understand it
if you have a stock engine (quest3d, vrml *shudder*), ask the developers or try different sizes/optimizations
oh and one thing: determine your target audience (i.e. graphics card + maybe the cpu) first
there are vast differences between the gc generations and chips (ati vs nvidia)
cu mike
It depends on your 3D viewer. Only some support mujltitexturing that allow you to use the same compositing technic than in Combustion After Effect etc…
In the exemple above I have render a *.vrimg from the vray frame buffer then extract the raw GI and raw lighting channel.
Then in the code for the real time 3D engine I use ( BSContact ) I have multiply the rawgi with the based material (diffuse) and then add the rawlighting.
If your 3D engine doesnt support multitexturing or multipath rendering your only way to go is to bake a complete material (diffuse x gi + lighting) wich will give a less detailed visual result.
It depends on your 3D viewer. Only some support mujltitexturing that allow you to use the same compositing technic than in Combustion After Effect etc
can I make this in max? I would like to use the baking method for animations with moving objects…
best regards
themaxxer
there is limited support for animation within vrml, especially using the blaxxun vrml exporter
that said, only object transformations (move/rotate/scale) are supported in a useful way. though by principle you could deform objects, the file size tends to explode if you do it. (each vertex position of the object gets saved for each animation frame)
i really haven’t looked into vrml lately but there are better realtime vizualisation systems out there.
those usually are external tools with their own authoring applications
some i know are:
quest3d
royalty-free, needs some programming/developer skills
rtre
http://www.cubicspace.com
cult3d
http://www.cult3d.com
you need to purchase a license to publish the projcet
virtools
be warned, though, all of these more or less need a high level of knowledge about realtime 3d and tend to need programming skills to give the spectacular results seen in the demos
you are wrong regarding VRML capabilities.
You are right : it requires some knowleges about realtime 3D.
Also you must know that the reference you quote : blaxxun viewer and exporter are outdated and havent been updated since 5 years…
Bitmanagement.de owns now and update frequently the viewer and the max exporter.
which VRML capabilities am i wrong about?
i know that vertex animation is possible, but as far as i know the verices’ position is sampled depending on your frame rate (f.e. 25 times per second) and saved to the file
as i said, haven’t used vrml for some time (last time i used it professionally was about 2000, when most people didn’t even have 3d accelerated graphics cards)
was just checking blaxxuns’ site and saw that contact (their viewer) was still online. i didnt know it was outdated
ouch, just checked bitmanagement’s prices for contact… if i remember correctly the viewer was free when blaxxun was offering it - the same goes for the max exporter which now requires the purchase of the sdk for at least 2000€
The price realy depends on who you are and what you are doing.
I know that this is a strange buisness technic but they work like that…
Just drop them a mail asking for the license package price, discribe exactlty what you want to do, and if it is not a one billion dollar project then maybe the license’s price could drop down
If you mention that Laurent PRIVAT (me) talked to you about the plugin exporter and SDK it could also help
Looks good n6.
I use Quest3d a lot, it is a fast DirectX 9 engine and is a node based environment so while some programming concepts are used like loops etc you don’t need to write a line of code. The cheapest version is 129 Euros and has all the functionality you need to do walk through .exes. There is also a supportive comunity based around the forum.
Nice example n6. You are quite right, the size of the map depends on the video card you are using. I’m using a Quadro FX 4400 so I’ve got 512 Mb of Video Ram the play with. Still not enough. The problem is my clientele are running crapping Pentium 4’s and even some P3’s that barely have 64Mb.
I forgot to mention in my last post that most of the projects we work on are very large scale. I’m talking about over 1 million polys in real time spanning over a couple of buildings. I’ve found that texture sizes seam to affect real time speed alot more than the amount of polys, but obviously still play a big part. I would go so far as to recommend vertex baking as a better option/combination. If you use Max’s radiosity engine bake in the solution. Then swap out what ever diffuse base material you like. Vertex baking replaces the need for a lightmap, although increases the need for more poly’s, but like I said before, I’m finding textures are heavier than poly’s. Find the medium ground. Doesn’t have to all or nothing.
On top of that, our real time stuff isn’t used for straight walkthough’s. We use EON Studio (www.eonreality.com) to produce content which has interactivity, like opening doors, moving objects, changing individual materials, animated objects (and animated meshes), so
a single lightmap over the whole scene doesn’t work for us, regardless of the size. Though I am interested in testing it having seen the posted example.
By the way, anyone producing realtime content for 3D Stereo display like we do, don’t forget video memory usage is doubled when going stereo. If you’re thinking about using 2048 or 4096 maps, then you really need the top of the line video card if your scene is actually more than just one room. Scenes with just floors, walls and ceilings only tend to be a bit deceptive when analysing video memory usuage. It’s kinda like using the Max teapot for testing. It all comes down to how big your scene is and how much detail there will be for final delivery. Also consider future stages of your project. If you use all the available video memory now, you’ll have nothing left if you need to add another, building or site to your project.
Although the other posts regarding larger maps sizes are true and achievable, if you have deadlines and client performance expectations to meet, I strongly stick by my suggestion to keep those textures as low as you can. Just look at what the games industry do with their stuff. If they can get away with using lower maps they will.
It’s all about efficiency and balance. Good to see so much discussion about baking. Keep it coming.
SunnyC
ivolve studios
@sunnyc1
I dont want to hurt you :lol: but a 1 milion polys scene isnt a real time 3D scene…
Realtime 3D walkthrough or fly involve a lot more of tech than only push the export button
Its like you where saying that to do a nice vray render, you only need to import the CAD model from Autocad and push RENDER :lol:
CAD models are allways full of useless polys and need to be cleaned.
But that is not all : serious realtime 3D walkthrough or fly or examine or what you want, require specifics optimization technics such as Level Of Detail, well organised BSP tree to do optimized culling, Normal Mapping to render a 100000 poly object only using 5000 polys and so one…
I could probably transform your 1 milion polys scene into a 100 000 poly one with exactly the same visual details and lighting quality (if not more ), the same interactivity and it would run very smoothly on a P3 with a Geforce2.
Regarding the stereo display its the geometry that you draw twice, the textures are instanced (if your engine is well programmed).
Regarding the video memory usage there are also a lot of specifics optimizations, one of them is to use DDS compressed textures wich is the native GPU texture format (but you engine must support it) and a lot of other…