Projecting A Texture Onto Geometry

[bradon]

Im trying to find out how to project a texture map across all the objects in the camera view. then bake the texture onto those objects so that when they move or when the camera moves the texture doesnt stay connected to the camera. ive tried this two different ways which ive outlined in these image files

http://www.paraportable.net/images/projection\_test.jpg

this works pretty well to do a planar mapping in UV menu then select camera. what is a projection but i need it to match the whole resolution gate instead of the bounding box of the objects. so i manually scaled the projected manipulator so that it was close to the resolution gate however its not exact and ill need this to be exact as the texture will be matching a lockoff footage from another shot. so the two are supposed to go together seamlessly.

1.origional texture and lockoff that needs to match.
2.projection settings
3.projected manipulator
4. bad render not matching
5. manually rescaled projection manipulator
6. closer matching render
7 animated objects

the problem is in step 5 i do the scaling manually and im afaid ill get it a little off and it wont have a seamless match.

IS THERE A WAY TO SCALE THAT MANIPULATOR TO THE RESOLUTION GATE EXACTLY???

my second test is here

http://www.paraportable.net/images/Camera\_...ection\_test.jpg

this time I mapped an "as projection" file texture node into the shader. well this gives the exact lockofff that i need to match but the texture doesnt "stick" to the objects when they move. I want to bake the image onto the geometry UV's or something so that when they move the image moves with it. i only need the first frame to be the match.

1. origional texture
2. hypergraph node
3. PERFECT MATCH image!
4. BAD! texture staying baked to the camera projection.

IS THERE A WAY TO BAKE THE TEXTURE ONTO THE GEOMETRY SO THAT THE TEXTURE FOLLOWS THE OBJECTS ONCE THEY START MOVING?

the goal is to get the exact projection match in the second example along with the way the texture sticks to the geometry in the first example.

[joojaa]

use a texture projection shading node (NOT a uv projection node wich something totay different wich you do here but dont want to do). Or sue background texture.

[bradon]

1.I did a search for texture projection shading node and all i got was how to make a UV projection node, so im not sure what you mean? possibly can you make a quick example or screen cap of what this node is and how its used?

2.Im not sure what you mean use background texture? You mean use background Material? I thought that was for catching shadows and reflection pass only without visibility, not sure how that relates to projecting a texture across all the objects in a scene. again perhaps an example of how this is executed would be very helpfull.

Thanks,
Bradon

QUOTE(Joojaa @ 06/07/07, 04:41 AM) [snapback]267670[/snapback]
use a texture projection shading node (NOT a uv projection node wich something totay different wich you do here but dont want to do). Or sue background texture.

[bradon]

Ok well the working solution for this is to use the second described method and to camera map the projected UV texture node, and then create a "texture reference object" for all the objects in the scene on the first frame, that holds the texture relative to that perspective and time, so when the objects move around the texture "sticks" to the surface.

maybe there is a better way, but thats what im going with right now.

[joojaa]

when you create a texture in the top theres a switch that says normal projection etc choose projection itll automatically hook a projection node to it.

And yes your not looking hard enough

[bradon]

QUOTE(Joojaa @ 06/07/07, 11:31 PM) [snapback]267731[/snapback]
when you create a texture in the top theres a switch that says normal projection etc choose projection itll automatically hook a projection node to it.

And yes your not looking hard enough

hmm well this seems like maybe you only read 1/2 my origional posted topic. because thats exactly what i think im doing in the second scenario there. the problem was how to get the texture to "stick" to the surface using the project texture node from the camera. because well its always projecting from the camera. and what i wanted to do was have the first frame project the texture onto all the objects, then somehow bake the texture onto the surface so when the objects or camera move, then the texture "stays" with the objects. so I think the way this is achieved was to do the project texture node as I tried in the second scenario of the origional post and as you mention here, but (and its a big but), then to make a texture reference object out of the whole scene geometry so that the texture stick from that angle and perspective. so say when the pieces move and break away then they look like they are taking part of the picture with them instead of looking like a projected light shining across them.

just to be sure we are talking about the same thing and so i dont go on another ghost hunt ill post another picture of what i am doing in the second scenario and you can tell me if thats what your refering to above

http://www.paraportable.net/images/projection.jpg

[joojaa]

Yes offcourse, it owrks on the space. What you would have wanted is uv projection instead then! But offcourse reference works quite fine too its the same thing just 2 times heavier.

[bradon]

well i guess i could stop there because the reference object does work, HOWEVER someone once told me to do things because they were smart not because they are easy! and well i think that creating a reference object is easy its not necessarily the smartest way, because your right it makes it 2 times heavier. And in certain cases it seems better to have the UV map so if i needed to paint precise details on the geometry i could do that. So then how would i create UV's based on the camera SO THAT THE ENTIRE UV SPACE IS DIRECTLY PROJECTED ONTO THE GEOMETRY AND SCALED TO THE RESOLUTION GATE AND NOT TO THE BOUNDING BOX OF THE OBJECTS SELECTED IN THE SCENE? It seems like when i do a planar UV mapping, then this is the only option i get which gives me a projection that stretches the image to the bounding box of the objects, ( see link A ) BUT I want to have the image Project out and stay at the same size as the resolution gate. NOW I Could just go ahead and stretch the projection manipulator, ( see link B ) HOWEVER I want this to be and exact match as best it can be. and my screen resolution only affords and eyeball prescision of so so quality.

so what is the proper way to get a UV projection that maps the image across all the objects and matches the UV space to the resolution gate?

Link A
http://www.paraportable.net/images/ProjectUV.jpg

Link B
http://www.paraportable.net/images/ProjectUV\_notes.jpg

[joojaa]

Yes thats a bit unfortunate. heres 2 ways. i would invariably opt for the second one because i use it a lot anyways for all manner of funky projections, but mainly because its more automatable. im telling the first way only becasue i think it makes more sense for a manual worker.

way one Append a plane that fits the camera gate to the object, project measure points in plane and move the lower corner and all other points to 0,0 measure ditance to upper corner scale around origin so that that value becomes 1,1. Delete plane.

transform points to camera space do a plamnar projection along z (with right gate attribs but thats just trig so), youd probably do this on a copy and transfer the uvs and delete the copy, off course this is doable in mel quite easily but more of api way of thinking

[bradon]

boy oh boy, im not sure if this is a language thing or if its just me. I mean id be interested to know if anyone else here can follow this logic.

I mean i can understand most of the theory of what you are trying to say here. somehow get a plane in the scene that matches the film gate, then project out onto the plane as it is the exact size projection to match. then it gets a little fuzzy, im to delete the plane and somehow transfer the projected points over to the rest of the geometry?

but you totally lost me on the particular steps to do this. I need to understand the way that this is achieved. If i understood the particulars i guess i wouldnt be posting in the forums so here goes.

QUOTE(Joojaa @ 06/10/07, 11:19 AM) [snapback]267841[/snapback]
Append a plane

not sure if this is a technical term in maya? never herd of "appending" something. maybe parenting, constraining grouping even but not append?

QUOTE(Joojaa @ 06/10/07, 11:19 AM) [snapback]267841[/snapback]
project measure points in plane and move the lower corner and all other points to 0,0

can you be more descriptive of how to do this?
project a measure point from where? to where?
lower corner of plane? or lower corner of film gate?

QUOTE(Joojaa @ 06/10/07, 11:19 AM) [snapback]267841[/snapback]
measure ditance to upper corner

mesure the distance to the upper corner of what?
if this is the film gate, How do i measure the distance from the lower corner of the film gate to the upper corner?
It seems like the film gate is a virtual thing seen only through the camera so how to use a measure tool for something like that?

QUOTE(Joojaa @ 06/10/07, 11:19 AM) [snapback]267841[/snapback]
scale around origin so that that value becomes 1,1. Delete plane.

scale the plane?
If i've moved all the points in the plane down to 0,0 then i dont understand how scaling it up will have any effect at all. If all the points are 0,0 as you specified then if i scale it to 1000 they will still all remain at 0,0

QUOTE(Joojaa @ 06/10/07, 11:19 AM) [snapback]267841[/snapback]
transform points to camera space

transform which points to what camera space? and by what means? a null? a parent?

QUOTE(Joojaa @ 06/10/07, 11:19 AM) [snapback]267841[/snapback]
do a plamnar projection along z (with right gate attribs but thats just trig so), youd probably do this on a copy and transfer the uvs and delete the copy,

so im to project Planar UVs on a copy of what? all the geometry in the scene? and if i've already deleted the plane im not sure how this is going to give the right gate attributes.

[joojaa]

its not logic its a workaround. Ill post something later whan i get home.

QUOTE
transform which points to what camera space? and by what means? a null? a parent?

geometry. You may benfit greatly from this brief video that explains how a 3d camera works.

3DCameraWorksLikeThis.mov

watch it with time, and scrub till you get it.

Anyway dont mix the two last paragraphs together they are 2 different ways to approach same thing. Thats invaraibly my bvad because i didnt say so.

[dwgregg3d]

Now I'm faced with exactly the same problem.

I need to lock, bake, nail down or super-blue the camera projected texture image from a single camera to a UV set. The planar projections Maya offers simply do not do it always resulting in distortions, stretching and such. I looked for a script or plugin that would allow a texture from a perspective projection to be converted to a planar (UV) set of coordinates. Alas, don't see any way to bake the perspective projections to UV's from a given camera position at a given time.

I need to do this for making some highly realistic textures from digital photos and a lot of them so the process needs to be automated.

[joojaa]

Atleats the problem you describe is NOT the same.

QUOTE(dwgregg3d @ 07/14/08, 08:46 PM) [snapback]288326[/snapback]
The planar projections Maya offers simply do not do it always resulting in distortions, stretching and such. some highly realistic textures from digital photos and a lot of them so the process needs to be automated.

A uv projection is not disortion free, reason is that texture coordinates must be interpolated along triangles resulting in following kind of problems:

Example on how the interpolation of a polygon object happens PER trangle not per face, so the interpolygon uv interpolation is allways linear tto the uv edge. This si NOT by teh way a maya issue. its a technology limitation fo polygons themselves.

Hence you cannot do a perfect uv projection for all conceivable objects. Unless you fix the triangulation by triangulating the mode and have very fixed edge disortionsl. However it is best to actually use screen space projections for stuff that really needs to be accurate, not only are they more flexible but also more problem free. Since problems start to add up once you start moving individual points for some geometries.

[dwgregg3d]

Perfect is to my needs something that looks good enough to pass visual muster, not so much of an absolute in technical terms.

The whole projection map routine is a little odd as the most practical thing I've found is to create a duplicate camera (resisting my instinct to consider cameras as expensive items) and just lock its channels and leave it as a projector. It just feels a bit uncomfortable to have an object and camera placed that are vulnerable to being moved or have to be tied together if it has to move.

The "convert" shader to texture routine works, but from what I can see it is limited to making a 4k texture. I've got some very big landscapes of great detail to fly over and 4k may seem like a lot, but it sure seems small when there are so many details and too much distance to travel.

[joojaa]

QUOTE(dwgregg3d @ 07/16/08, 07:31 AM) [snapback]288398[/snapback]
Perfect is to my needs something that looks good enough to pass visual muster, not so much of an absolute in technical terms.

Yes but unfortunately uv mapping only works well for lineary placed uv distributions ACROSS the polygon, if you introduce polygons that have a perspective distortions than the triangle has to split teh polygon some way and the cordinates the triangle encounters will not look ok on both sides of the polygon triangle.

Its s inbuilt problem of how polygons work. Theres no way around it if you use POLYGONS. If youd used micro tessaletd subdivision surfaces then its an other deal. But youd pretty much need prman for that, as mr is not doing the right kind of dicing.

Try it with a l shaped polygon and make triangles then skew the uv map and calculate by hand how to solit is so the uv disortion never affects the interpolation. Cant be done its explained in the manual under the mantalray reference. (its not a mentalray issue but a polygon issue)

QUOTE
The whole projection map routine is a little odd as the most practical thing I've found is to create a duplicate camera (resisting my instinct to consider cameras as expensive items) and just lock its channels and leave it as a projector.

thats the only way thats going to do 100%! job for polys. You can allways hide the node from users and mark the node as reference then the likelyhood of somebody doing this by accident is allmost 0.

B the way nothing differes here form having a uv map even its not locked. And beyond change. And offcourse if you need to sirort the map you can allways use reference geometry.

QUOTE
The "convert" shader to texture routine works, but from what I can see it is limited to making a 4k texture.

Its not limitted to 4k, just the gui goes no further but theres NO such limitation no. Also mr baking can do as buig maps as youd ever want out of the box. But bacause it performs same thing faster i see this as nobrainer issue. Tough you can make it EVEN faster with mr.
QUOTE
I've got some very big landscapes of great detail to fly over and 4k may seem like a lot, but it sure seems small when there are so many details and too much distance to travel.

Texture map size is one of the most misunderstood things. People have a tendency to over exagarate the needed texture resolution. On the avarage scene sent for me to optimize the resolution of textures is in general 2-8 times too large. The most prominent example being the blue marble maps wich people want to use a whole.

For starters a too big map slows renders down, because more filtering needs to get done. But allso quite supprisingly lower res maps often produce BETTER images than too high res maps. This is especially true for scenes that use motion blur, and fast movement where this actually speeds rendering by heaps.

In practice I never need to go beyond 8k textures for 2k production EVEN with full world zoom in zoom out, zip from place to place. When people go big in their maps they often do use the texels of highest res but nly for a small small section of the map so why not just have a locally more accurate map where they smack on need it?

PS: the uv problem can be aleviated tough, but in practice this results in surface normal problems instead so its not recomended. Ill post pictures at some point.