Shading based on normals

[djsmackmackey]

ok, Here's the final goal I'm working on. (I hate giving things away before I've perfected them )I'm working on a way to use an image as a form of lighting. kind of a fake HDRI. so, I use the envBall or envSphere to map the image, but instead of putting it on an object, I put it on a light. so the light projects the image. with careful light placement and adjustments, I can match the lighting of a live action plate with this light probe. Now the bad part. see it's just a fake. if I move the camera, the image is actually being reflected, er , sort of, so it starts shifting around like a reflection. very ugly. but the catch 22 is that only an envBall or envSphere can project the image acording to the normals. I've been working for 2 years on this, and it's this same blocking point. so I recently figured out I can use photoshop's polar distortion to make a map that I can then bring into Maya, connect the sampler info node's facing ratio to the V coordinate of the image, and I get the same result. the normals dictate where the image is placed, not the objects uv's. So I figured there must be some way to change the facing ratio to read from a locator or something like that, as opposed to the camera. kinda reading the locator as the camera and shading it that way. so instead of that radiating out from where the normals are parallel to perpendicular to the camera, it would be to the locator. still with me? Now, if I can figure this last little thing out, I'm all set (I hope). but until then, I can only use stationary cameras. IF you'd like me to write up a quick tutorial on this, just let me know, either on here or email at hireme3d@yahoo.com. Just think. HDRI type renders without needing HDRI technology, and not having to spend days and weeks making the lighting match just right. you could match lighting in just a few short hours.

[ggray7]

I believe something like this was shown at Siggy, maybe by Debevek and team. Remember they where getting light placement and intensity and color based on the hdri. Maybe a feature of HRDIshop? I am full of maybes today Looking forward to getting in front of maya so I can look into things and hopefully be a bit more helpful.

What I am curious about is why you want to have lights project the image compared to standard methods of hdri

~Garret

[joojaa]

So What you are basicaly triing to do is apply the color of the corresponding normals on each surf so that shadows would be taken into account...

Couldnt this be easily done with raytracing (but thats obiviously not what you want), with some nifty shaders and no lights? But 1 ray raytracing shouldnt be that hard or expensive.

What you could do is a simple vector space transformation. Since you fnow the position of your camera in world space and you know that the normals are in camera space then just re compute the camera space normals to worldspace normals. Now you could pluck shomewhere in the calculation that you expect the normals to behave as if the camera is in position xyz looking at point (you know the differennce of theese).... Read a math textbook on vectors and youll get the picture. Theres absolutely nothing there that hinders this.

But i think i seem to remember that you can get this info without so much hustle (maybe it was the api or something), ill investigate...

But im kindof busy right now so...

[ggray7]

Shading based on a definable axis compared to the surface normal(slope shading) for any surface including subD:

The idea behind this it we want to compare the surface normal of an object being shaded in world space, realtive to an axis, mostl likely the Y. (Slope for things like mnts and snow and so on) We want something to return a 0-1 based on an angle of 0-90 from the Y. This is fed to a ramp which can drive anything.

Steps:

Create a directional light to define the axis to compare to. (This is not really needed and could be replaced by numerical values in the next node, but the light allows for easy manipulation if you are unsure of what axis you want to compare againts. By axis I mean vector)
Obviously make the lights intensity 0 and unlink it to all.

Create a Vector Product Node and set it to dot product and normalize output.
Connect the DirectionalLight.lightDirection to the input1 of the vector node.

Create a sampler info node.
Conenct the normalCamera of the sampler info to the input 2 of the vector node.

Connect vectorNode.output to an RGB to Lum node(so things look nicer "code wise" )

Conenct the outValue of the RGBLum node to the X (or any) input of a set range. Min 0 Max 1 Old Min 0 Old Max 1. Not sure that is logical but it works.

Connect outValue X of set range to the V of a ramp and feed that to a shader.

Rotate your Dircetional Light to -90 and you'll have shading comparing the surf norm to the Y axis, assuming Y up.

Not sure everything is logical or most efficient as I made this a while ago and just found it at 3am. But if it helps anyone, great

I've tested it a decent amount and it works.

~Garret Gray

[djsmackmackey]

Oh Man,I think you got it!!!!!!!!! that is soooooooo COOOOOOOLL!!!!! I'm at work and can't check for sure, but I think you figured out the secret missing puzzle piece! I can't wait to go home and try it out. Damn if I didn't have bulls tickets tonight! I'll be up late tonight trying it out I have a feeling. How did you figure it out?? When I tst it out, if it works, I'll write up a tutorial and show you guys how it works to fake HDRI to match a background plate. If you haven't already figured it out yet. Talk to you guys later!!

Sean Skube

[hell-raiser]

Nice... Real nice...

One question.. What the heck is that set Range doing? I know it works but I cant wrap my brain around what its doing.

dave

[stevesayer]

Ahaha, awesome stuff... good work, GGray7!

I don't know about that SetRange of yours, it might be superfluous as it stands... its acting more as a clamp than a real SetRange, I believe. I could be wrong.

Thanks again for the tips!

-Steve

[ggray7]

I came up with this method almost a year ago with the help of discussions like this one, the cool people of highend, and #maya on irc which I need to get back to some day. I grabbed a cd of old work and found it late last night. At the time I did the shader I was less clear on vector math then now, so looking forward to giving it another real look.

btw
http://chortle.ccsu.edu/VectorLessons/vectorIndex.html

I have nothing to do with the site, but real nice explinations of things that directly relate to this topic.. like the dot ptoduct. Especially for all those self tought in CG math like me.

Like I said, the dir light is just one way to get the vector, nice cause you can rot it around intuitively if needed.

I took out the set range last night, but it messed things up, but it was 3am, so might have just been sleepy, need to take a serious look at it

So assuming that solved your question, can I now lead the topic to curvature shading on things other then nurbs? (still interested in what you are doing hdri wise with this too) Seems we have the attention of the minds who'd know here so maybe we can get some info on that for all to enjoy.

I've heard it can be done with nurbs, how bout polys? Curvature is based on the derivative of the normals surrounding the pixel being shaded right? So what methods can be used to compute that? For maya and in general. Assuming the need to sample points around main point of interest, how does maya accomplish that? Donno what we'll get on that topic as I've asked it before, but thanx to all who respond.

~Garret

[stevesayer]

Well, polygons really present a problem for this, don't they? --Because of course they are a collection of more-or-less flat shapes, and don't have a proper 'curvature' to them at all.

Anyone know how the 'smooth shading' works across poly surfaces, anyway? --I can see Jim Blinn's smiling face over there in the right-hand column, and I suppose he must know. But seriously... how DO renderers work with the normals of polys? For smooth shading, and bump mapping and such? Do they

Therein, I think, will lie the key to a method for polys...

SubDs I understand so little (mthematically) that I'm not even going to touch the subject.

-Steve

[ggray7]

Yes it'd make little sense to do curvature on flat shaded polys, and even on smooth shaded unless it's real dense. As far as I know, smooth shading is just intperilating the normals of each vertex using a standard method of doing so. Don't know what that method is off the top of my head, but seems pretty simple to grasp. It intpers the normals and shades accordingly. Bumps take the normal given by the smoothing algo and change it a bit.

SubDs are interpilation, subdivision, of the poly cage using some smart rules of how new vertices are created based on the valance(number of surrounding vertices). You can look at it as given a vertex, the number of edges connected to it is the valence. 4 would be called even, anything else is odd and different rules apply. All these sub algos carry over to the shaders you apply. SubD in maya subdivides down 4 levels by default(which is usually more then needed) and then renders that. So I assume it's rendering polys at that point as far as the renderer is concerned as I think all surfaces (nurbs being the addition) convert to polys at render time in maya.

Anways, for curvature you need to look at the normals of the points around the point being shaded and compute a rate of change of the normals. At least that's what I'd think is done. So the question then is how does one go about that in general, and in maya? Specifically is there a node that allows you to compute other points then the one you are on for any surface type(maybe one node for each type, whatever). How are issues like recomputing info and handling edges delt with? How far out do you have to sample? etc.

~Garret

[djsmackmackey]

Ok, I did a rough test here at work and I think you definitely solved it. I'll know more tonight when I get home and fiddle with it, but I do believe it will work now. I'm stoked. I promise, tonight, if it works, I will write up the process I've been working on. just don't get too excited, cause remember, it is a fake. not true HDRI. but I am hoping you will all be able to benefit from it, cause you have helped me so much with this. I had almost given up. Thanks guys.

[djsmackmackey]

Ok, well when I got home and tried it out it wasn't working the way I'd hoped. when I used an image instead of a ramp, the image either was just turned into concentric rings, or it was mirrored on the other side of the object. However, I did write up a tutorial and I put it online for you to read. Then you can see what I'm trying to do and maybe help me solve the problem. My tutorial is at http://members.aol.com/xeroacid77/fakingHDRI.htm
feel free to email me at hireme3D@yahoo.com and tell me what you think and maybe how to solve my dilemma. Cheers!

Sean Skube

[stevesayer]

The dot product is a scalar product, not a vector product. That means that it can only give you a single value for each sampled point (like FacingRatio does), not a triple value (like NormalizedNormal does). With only one value for each point, there's no distinguishing between points on the object that have normals which are on opposite sides of the control vector (your light), each facing away from it the same amount. This is where the concentric rings come from.

You might be able to play with the CrossProduct option in your vector multiplication node--the cross product is a vector product, not a scalar, so it will output triple values (a vector) for each sampled point. This might let you achieve what you're looking for.

I echo those who have said 'get a math textbook for reference,' though... the way the cross-product works is not really straightforward or intuitive if you don't know vector math.

Good luck!

-Steve

[moneyhouse]

just thought you guys might be interested, this is the same rendering technique ILM used for pearl harbor, and jurrasic park 3, if anyone is interested, check out this project im working on with the same type of technique http://www.kwinter.com/chas/b25.jpghttp://www.kwinter.com/chas/b25.jpg I used a couple of snow textures verry carfully

-charlie

[djsmackmackey]

Actually, I read that they used actual HDRI for those flicks. But they said they developed the process. My method is just a fake to get a similar lighting effect, although the shadows still have to be "hand-made". Are you from Jacksonville? on your website you mention the Dames Point Bridge. I used to live in Jacksonville myself.

[moneyhouse]

i read that they baked radiosity and then used the other technique on top of that in cinefex, anyway, im not from florida, but i though that bridge was cool so i used it as a setting in that video

-charlie

[djsmackmackey]

Oh, I only read about it in Computer Graphics World. It didn't go into detail, just said they used a mirrored ball to gather light data, which made it sound like they were just doing ordinary HDRI. As for my method, I need to find a way to make it work less like a reflection, and more like a normals shading texture. Maybe I should just give up. by the time I figure it out, they'll have HDRI standard in Maya. I could always try to find a way to make it an animated texture, so depending on the angle, it shows a different frame. but that's more trouble than it's worth I think.

[djsmackmackey]

Hey, I just wanted to thank you for your help. I finally figured it out, thanks to you. Everyone has helped in the best way. I'm going to go home tonight and begin revising my tutorial to show how you can fake HDRI's style of image based lighting to match the light in a live action scene. if any of you haven't seen my tutorial, please feel free to look at it at http://members.aol.com/xeroacid77/fakingHDRI.htm
and tell me what you think. It should be up in a couple days with the revisions. but the old version is still viewable. Also, it's in the VIP lounge at 3dcafe.com
Thanks again you guys!