Hi! I'm very new to OpenGL, just started learning it for a class at the university. I've already tried doing some OpenGL tutorials for the most basic stuff, and now I'm supposed to learn how to do shadow mapping. When I googled for it, I found this: http://www.paulsprojects.net/tutorials/smt/smt.html which seems to be just what I need, so I've decided to try it. (I'm not actually copying everything there, because I'm trying to do shadow mapping for my own simple scene instead of the scene from the tutorial, and I'm using QT which makes some of the code a little different, but I don't think this is the cause for the problem I have) When I started doing that tutorial, I found out that my compiler has absolutely no idea what the "MATRIX4x4" and "VECTOR3D" types are. So, I thought about using some different types... instead of using VECTOR3D objects I simply used normal integer numbers as coordinates in the functions which needed them. And, instead of using MATRIX4x4 I used [16] arrays of GLfloat. The glLoadMatrixf function, which in the tutorial uses MATRIX4x4 objects as its parameter, works well with using the arrays of GLfloat instead. (I even found some documentation online which says that the function is actually supposed to use GLfloat* as its parameter and didn't mention anything about MATRIX4x4). It compiles and runs without any errors. However, I got stuck at this line of the tutorial: MATRIX4X4 textureMatrix=biasMatrix*lightProjectionMatrix*lightViewMatrix; Unfortunately, here I can't use arrays of GLfloat, because they can't be multiplied in the way that matrices can be multiplied. And I can't really get around that, since this matrix multiplication is the essential part of shadow mapping... So, my question is: Since my compiler has no idea what the MATRIX4x4 type is, is there some other way I could do matrix multiplication? Or maybe there is some different type to store matrices? I'm using QT 4 on Linux...

Just download the source (linked at the bottom of the tutorial), and you'll find that these classes are included therein.

FYI, objects of the MATRIX4x4 class type can be used as arguments to (e.g.) glMultMatrixf() because the class supports implicit conversion to float*.

Quote:Original post by jyk
Just download the source (linked at the bottom of the tutorial), and you'll find that these classes are included therein.

FYI, objects of the MATRIX4x4 class type can be used as arguments to (e.g.) glMultMatrixf() because the class supports implicit conversion to float*.

Thank you! :) I haven't thought about downloading the tutorial source code...

Yes, there is. Basically the code is doing

glMatrixMode(GL_TEXTURE);
glLoadIdentity();
glTranslatef(0.5f,0.5f,0.5f+bias); //bias isn't necessarily needed and isn't included in tutorial
glScalef(0.5,0.5f,0.5f);
glMultMatrixf(projMatrix);
glMultMatrixf(viewMatrix);

But yes you can download the source code from the tutorial. Bare in mind, that this tutorial does not support Shadow Maps greater than the size of your monitor, which can cause artefacts. Since your using QT4, you may want to modify the code to use QGLPixelBuffer or QGLFrameBufferObject for better results.

Jamie

Quote:Original post by Neutrinohunter
Yes, there is. Basically the code is doing

glMatrixMode(GL_TEXTURE);
glLoadIdentity();
glTranslatef(0.5f,0.5f,0.5f+bias); //bias isn't necessarily needed and isn't included in tutorial
glScalef(0.5,0.5f,0.5f);
glMultMatrixf(projMatrix);
glMultMatrixf(viewMatrix);

But yes you can download the source code from the tutorial. Bare in mind, that this tutorial does not support Shadow Maps greater than the size of your monitor, which can cause artefacts. Since your using QT4, you may want to modify the code to use QGLPixelBuffer or QGLFrameBufferObject for better results.

Jamie

Thanks Jamie :) I'm actually new to QT as well (I started learning both QT and OpenGL at the same time) so I'll need to look around the documentation for some information how to use these QGL objects...

I should add that you need to put a
glMatrixMode(GL_MODELVIEW);

after the code I posted, otherwise any computations would still be on the texture matrix rather than the modelview or projection matrices you probably use.

Well for QT Documentation the Tools which come with the download are pretty good. As for any SM knowledge, this board has about five-ten topics at least in its subsequent pages. Most of them are probably mine and Matthias' :)

Jamie

Hi! It's me again, thank you for the help, but I need to ask some more questions because I still have one problem, since what I did isn't working :(

EDIT: It ALMOST works... I have a rotating cube above a plane, and there's a light high above them, so that the cube is supposed to cast a shadow on the plane. And it does... except that it's totally wrong. Here are some screenshots which show how wrong it is:

http://i11.photobucket.com/albums/a156/Miracle86/bad-opengl1.png

http://i11.photobucket.com/albums/a156/Miracle86/bad-opengl2.png

http://i11.photobucket.com/albums/a156/Miracle86/bad-opengl3.png

I've tried playing around with the settings, and I've found out that changing the last two values in this line (which defines the light projection matrix) so that the near clipping distance and far clipping distance change, then the whole "shadow" changes as well. And also for some values the shadow disappears totally, and for some other it covers the whole scene...

gluPerspective(45.0f, 1.0f, 2.0f, 8.0f);

However, even if I pick the near and far clipping distance so that it covers the cube and the plane below it, the shadow is never actually correct. It appears to move in harmony with the cube's rotation, though... does anyone have any ideas what can be wrong in my OpenGL now? :)

[Edited by - Starshine_ on April 4, 2008 11:01:16 AM]

It looks to me like you have shadow leakage and projection artefacts.

Try a near plane of 1.0, far plane of around 150 and see if that helps. You may also need to move the light back further slightly. Though I can't tell without some more views or movement around the application.

Thanks for the answer, I've tried doing it, but when I do it then the whole scene is shadowed. :( It seems to be the best when the near plane of 7.0 and the far plane of something much higher, but even then it's far from perfect, kind of like on the screenshots I've posted.

Unfortunately, my OpenGL program is really simple and doesn't allow movement... but tomorrow I can try to record the animation on my computer and upload it somewhere. And also here are some facts about the position and size of the elements of the scene...

The cube's edge length is 2, and its center's position is 0.5, 0.0, -6.0. The plane below it is square, its width is 20... and its Y position is -1.8. There is one light, and its position is 0.0, 10.0,-3.0... so it's quite high above the objects. Should it be even higher?


I also have one idea, maybe it all is a result of my imperfect C++ skills... since I'm using an array of GLfloat to store matrices instead of the MATRIX4X4 objects from the tutorial, I had to do some things differently... for example, this line from the tutorial:

glTexGenfv(GL_T, GL_EYE_PLANE, textureMatrix.GetRow(1));

(This is the second row of the matrix, the first one has the number 0 in the tutorial)

looks like this in my program:

glTexGenfv(GL_T, GL_EYE_PLANE, &textureMatrix[4]);

Is that a correct way to give the glTexGenfv function a specific row of the matrix?

EDIT: I know what I did wrong! At least I think so... I've just looked around some OpenGL documentation, and it seems that the values from matrix[0] to matrix[3] are the first column of a matrix, not the first row...

EDIT 2: Fixing that didn't help... but I've noticed another weird thing. In the init function, before saving the camera view matrix, gluLookAt is used... but no matter what the first three parameters I give it are, the camera position is always the same, at the center of the scene... maybe for some reason that part of the code doesn't generate or save the correct matrices for the light and for the camera?

EDIT 3: Another update! When doing the matrix loading and multiplication stuff on the GL_TEXTURE matrix, I decided to include a glPushMatrix() and glPopMatrix() before and after doing everything, so that doing these things doesn't change whole matrix. And now it's much better! :D The shadow seems to be generally displayed correctly... its shape and animation looks good, the thing that's wrong is that the shadow is bigger than it should be... well, I'll try to look around the program and see what I can do to change it. :) But I still haven't found out why changing the camera view matrix by the gluLookAt function doesn't actually change the camera position... (UPDATE: the shadow's shape still isn't correct, because a lot of glitches randomly appear, depending on some things, such as where exactly the light is pointing at according to the second row of the gluLookAt function... especially the cube is shadowed almost all the time, even when the shadow on the floor plane looks mostly correct... and I've just noticed another thing. The shape of the shadow is exactly the same as the shape of the cube, but from the camera's point of view, not from the light's point of view... but the place where the shadow on the floor appears is correct, because it's under it, and the light is high above it)

[Edited by - Starshine_ on April 6, 2008 8:14:49 AM]

Bigger than it should be? Mmm. Post a screenshot.