PDA

View Full Version : About UV-s and Projections



allankiipli
03-08-2004, 01:37 PM
I have been thinking about quick and natural way of texturing.
We have possibility in Lightwave to project images as planar
mappings to objects.

What i want, is the possibility to do projections only for polygons
that face the projection direction, and to project in all six directions: xyz and negative xyz. This makes possible to take
othogonal screenshots in modeler and project them in cubical
manner to the object. It could elliminate the need to make uv-s
at all.
This could be a special cubical mapping that takes six images
and blends pixels from image pairs.

This sort of workflow has advantage that the painting is more
predictable and natural instead of squeezed uv layouts.

mkiii
03-08-2004, 05:23 PM
What you would end up with is a very streaky texture where 2 projections overlapped.

You have always been able to map this way using planar projection. The use of alpha masks around the edges of the various textures will blend each into the other. This works fine for complicated shapes like 4 legged animals, and things that don't have a distinctive pattern that needs to match up exactly.

You would still get more control using uv maps - which don't have to be squished btw.

Instead of creating the uv coordinates in auto mode when you first make the UV map with the 'T' dropdown menu, deselct the Initial Valuse button, select the polys you want in a particular view, and use the texture guide numeric panel, with default settings & Make UVs selected. The resulting UVs are completely undistorted, and are a no brainer if you create a square texture map to paint on (use eps export to get your uv map out).

Now grab the new UV,s and move them aside, & repeat for the other views.

Once you have all the sides, just arrange em around the UV grid or copy each one to a different UV map. I use this method for just about everything - so far no complaints & the only work you need to do is to drag out the bordering edges on parts that are very rounded.

allankiipli
03-09-2004, 02:07 AM
I know the ways.
But this sort of mapping might be extended to automatic setup.

Technically with uv maps the pixels of a map become always
distorted unless we map a object like a box.

With this method, the computer uses a 100% normalprojected
polygons: It dissassembles the object and creates info at
given resolution from three, two or one points. Depending of
the orientation of the polygon:(it can face maximum three planes
at the same time) The facing amount determines the amount of blend. (Actually not image pairs but three images.)
This means:
no new technology must be integrated. It still has one image
source for pixel information. But image and uv-s are generated.

Maybe i should try to program something myself--- too complicated i think.

Karmacop
03-09-2004, 02:33 AM
You can already do this, just use 6 planar projections, but it will turn out crap. That's why people use uv maps instead of 6 planar maps.

allankiipli
03-09-2004, 04:35 AM
you cannot have blending as described above.
And this is a tedious setup.
The reason for this is speed and accuracy.

mkiii
03-09-2004, 05:19 AM
You can have blending as I said earlier, with alpha maps to soften the edges of each projection. This is how it was done in LW before V6.0 introduced UV mapping.

How on earth do you expect to be able to automatically map an object in this way & not have extremely stretched textures where 2 projections meet?

Karmacop
03-09-2004, 06:40 AM
For the blending either use hand painted alpha maps or you can easily set up gradients.

The reason you'd map anything like this is for speed and ease only. You'll get horrible texture stretching and more importantly, overlapping mapping.

If you were going to do it this way I'd paint my texture like this, then baked my textures to a uv map and fix them up there. I'd never want this for a final render though.

allankiipli
03-09-2004, 07:06 AM
You cannot get the point.

Accuracy means no stretching at all.

What im talking about is to use three times more accuracy than
before. If a pixel gets defined in three sources instead of one,
he is three times more accurate.

You only have to paint what you expect to see. And it has a
nice combination in output: No UV-seam like thingy.

Alpha could be generated with camera incidence maybe, but we
dont have orthogonal camera in lightwave.

And one more time: the projections work currently only in three
directions. This means: If i use front image. I get this projected
also onto back.

Im not against UV-s. But im trying to explain true cubic mapping.

allankiipli
03-09-2004, 07:19 AM
with alpha generating i go home and look at this

Karmacop
03-09-2004, 08:25 AM
Believe me, doing things this way is not more accurate.

As for alpha, I'd use light incidence and have a light on each of the 6 sides of the object.

If all you want is cubic mapping where you get to have 6 different images then ask for it ... email [email protected] and tell them. But this is a bad example of a model to use it on.

mkiii
03-09-2004, 11:04 AM
He's right. I cannot get the point.

allankiipli
03-09-2004, 12:10 PM
Advantages of true cubic mapping (might be) :

*No UV seams
*Fast Workflow
*Possibility to get exactly what you paint. (Its easyer to work
with orthogonal views of object than approximating from distorted
UV-mesh. )
*Advantage to project texture as photographs from real objects with no big modification.
*Uniform distribution of pixel data: little polygons get less info
large polygons get more info.

I see no reason why it should look like crap.
The flow of info is straightforward.

Of course the problem is with more away-facing polygons, where
you have not inasmuch control than with facing ones. But remember, that human eye
thinks in terms of front left top projections also and forgets
inbetweens.:)

allankiipli
03-09-2004, 02:46 PM
i tryed the light incidence, and it really works and elliminates
backside projections.

Result is far from crap. Actually good result with a crap texture.

Problem is that the lights can be used only with static objects.
When i begin to deform with bones, alpha blendings change.

allankiipli
03-09-2004, 02:50 PM
result

mkiii
03-10-2004, 01:44 AM
Now try the same technique with a pattern such as polkadots of a chequerboard.

All you have there is a purple hair tex blurring into a pink head, How is that an accurate mapping technique. Obviously, the flesh tones, will blend correctly - they are all the same colour and contain no detail.


I'm telling you as a fact that there is NO way that you can map anything more than a cube or a sphere like this without either getting smearing, blurring or stretching, or a hard seam where the 2 planes meet, depending on the technique you use.

The amount of work you have to do to correct this sort of thing, far outweighs the amount of time it takes to set up a decent UV map.

allankiipli
03-10-2004, 12:15 PM
I think you are not right.

The way is decent.

Notice how eyebrow blends in the outer end.

The real beauty is ACTUALLY taking place where the blend is.

And of course im not going to paint hair texture onto bald
head. This is a test.

The only problem with my current model is, that i cannot take
large screenshots enough. This should be programmed in.

I imagine also sort of paint application, that might work with
these projections. To paint straight to model what has no UV-s. This might be fun. Could be done.

For now i must probably bake the texture layers and still use UV vertexes.

allankiipli
03-11-2004, 03:49 AM
As we see the problem areas are the surface units who duplicate
themselves in means of their orientation: as the inside of breast.

This could be internally solved probably via some depth information.

allankiipli
03-11-2004, 04:11 AM
And the screenshot of what was set up

allankiipli
03-11-2004, 04:13 AM
:D

Karmacop
03-11-2004, 07:24 AM
It'd good to see you're working on this man :)

But yes, as we were explaining, you run into a lot of problems as the models become more complex.

allankiipli
03-11-2004, 08:04 AM
Of course a beast with 8 hands would be probably more problematic to map this way.

But i was thinking the resolution of the projectionbox might be
lifted. You could be probably to be able to insert divisions and
thus lift the number of projective image sections.

But this is probably not the right way to do it: too much waste
of pixel data. Maybe the large emty areas might become compressed out in some manner.

Im i right that for now i must render with surface baker separate
passes to get those lightgenerated alpha channels.
But then the backside projection comes again onto surface.:rolleyes: ?

allankiipli
03-21-2004, 08:56 AM
Rendered via faked True Cubic.
Same map applyed to Bump.
Info:
Front Projection: 1256*1327
Back: same
Right: 244*1328
Left: same
Top: 1256*250
Bottom: same

Image Pixels
4 609 488

Images screenshots from modeler: sketch mode, combined up
via pieces, unscaled.

allankiipli
03-21-2004, 08:59 AM
Baked to Atlas UV-s
1024*1024

allankiipli
03-21-2004, 09:04 AM
Baked to Atlas UV-s
2147*2147

4609609 pixels

JDaniel
03-21-2004, 11:57 AM
Originally posted by allankiipli
I think you are not right.

The only problem with my current model is, that i cannot take
large screenshots enough. This should be programmed in.

I imagine also sort of paint application, that might work with
these projections. To paint straight to model what has no UV-s. This might be fun. Could be done.

For now i must probably bake the texture layers and still use UV vertexes.

Allan can you do that again w/ a small grid pattern?
Size up your screenshots in PS to desired res. you want before you start.
You're talking about projection painting. 3d paint apps. do this already and don't project through other side.
The closest in LW to that, is how you're doing it w/ incident alphas on planar projections.(which limit you to 3 axis) Or my crazy fixed projections tech. Both need to be baked to a final image too. And you have to have uv's as you know to catch your texture.

allankiipli
03-21-2004, 03:53 PM
The two last pics are baked to the atlas mapping who looks like
this: (scaled down 50%)

allankiipli
03-21-2004, 03:58 PM
and with checker pattern

allankiipli
03-21-2004, 04:14 PM
closeup

JDaniel
03-21-2004, 08:01 PM
Not bad, but it looks a little distorted where the alphas meet.

Karmacop
03-21-2004, 09:23 PM
Yeah it gets distorted where the alpha meets. The texture can actually get stretched 140% or so in each axis. At the most distorted point, even though you have 3 pixels, they'll all be stretched the 140% .. and there's nothing you can do about it using this method.

JDaniel
03-21-2004, 11:55 PM
This is where you need an algorithm to fix.

Karmacop
03-22-2004, 01:36 AM
Algorithm? It's the nature of how he's mapping it. If he wants less distortion he should use a uv map :rolleyes:

allankiipli
03-22-2004, 11:53 AM
Thats it, but its still in consideration.