You can subscribe to this list here.
2000 
_{Jan}

_{Feb}

_{Mar}

_{Apr}

_{May}

_{Jun}

_{Jul}
(390) 
_{Aug}
(767) 
_{Sep}
(940) 
_{Oct}
(964) 
_{Nov}
(819) 
_{Dec}
(762) 

2001 
_{Jan}
(680) 
_{Feb}
(1075) 
_{Mar}
(954) 
_{Apr}
(595) 
_{May}
(725) 
_{Jun}
(868) 
_{Jul}
(678) 
_{Aug}
(785) 
_{Sep}
(410) 
_{Oct}
(395) 
_{Nov}
(374) 
_{Dec}
(419) 
2002 
_{Jan}
(699) 
_{Feb}
(501) 
_{Mar}
(311) 
_{Apr}
(334) 
_{May}
(501) 
_{Jun}
(507) 
_{Jul}
(441) 
_{Aug}
(395) 
_{Sep}
(540) 
_{Oct}
(416) 
_{Nov}
(369) 
_{Dec}
(373) 
2003 
_{Jan}
(514) 
_{Feb}
(488) 
_{Mar}
(396) 
_{Apr}
(624) 
_{May}
(590) 
_{Jun}
(562) 
_{Jul}
(546) 
_{Aug}
(463) 
_{Sep}
(389) 
_{Oct}
(399) 
_{Nov}
(333) 
_{Dec}
(449) 
2004 
_{Jan}
(317) 
_{Feb}
(395) 
_{Mar}
(136) 
_{Apr}
(338) 
_{May}
(488) 
_{Jun}
(306) 
_{Jul}
(266) 
_{Aug}
(424) 
_{Sep}
(502) 
_{Oct}
(170) 
_{Nov}
(170) 
_{Dec}
(134) 
2005 
_{Jan}
(249) 
_{Feb}
(109) 
_{Mar}
(119) 
_{Apr}
(282) 
_{May}
(82) 
_{Jun}
(113) 
_{Jul}
(56) 
_{Aug}
(160) 
_{Sep}
(89) 
_{Oct}
(98) 
_{Nov}
(237) 
_{Dec}
(297) 
2006 
_{Jan}
(151) 
_{Feb}
(250) 
_{Mar}
(222) 
_{Apr}
(147) 
_{May}
(266) 
_{Jun}
(313) 
_{Jul}
(367) 
_{Aug}
(135) 
_{Sep}
(108) 
_{Oct}
(110) 
_{Nov}
(220) 
_{Dec}
(47) 
2007 
_{Jan}
(133) 
_{Feb}
(144) 
_{Mar}
(247) 
_{Apr}
(191) 
_{May}
(191) 
_{Jun}
(171) 
_{Jul}
(160) 
_{Aug}
(51) 
_{Sep}
(125) 
_{Oct}
(115) 
_{Nov}
(78) 
_{Dec}
(67) 
2008 
_{Jan}
(165) 
_{Feb}
(37) 
_{Mar}
(130) 
_{Apr}
(111) 
_{May}
(91) 
_{Jun}
(142) 
_{Jul}
(54) 
_{Aug}
(104) 
_{Sep}
(89) 
_{Oct}
(87) 
_{Nov}
(44) 
_{Dec}
(54) 
2009 
_{Jan}
(283) 
_{Feb}
(113) 
_{Mar}
(154) 
_{Apr}
(395) 
_{May}
(62) 
_{Jun}
(48) 
_{Jul}
(52) 
_{Aug}
(54) 
_{Sep}
(131) 
_{Oct}
(29) 
_{Nov}
(32) 
_{Dec}
(37) 
2010 
_{Jan}
(34) 
_{Feb}
(36) 
_{Mar}
(40) 
_{Apr}
(23) 
_{May}
(38) 
_{Jun}
(34) 
_{Jul}
(36) 
_{Aug}
(27) 
_{Sep}
(9) 
_{Oct}
(18) 
_{Nov}
(25) 
_{Dec}

2011 
_{Jan}
(1) 
_{Feb}
(14) 
_{Mar}
(1) 
_{Apr}
(5) 
_{May}
(1) 
_{Jun}

_{Jul}

_{Aug}
(37) 
_{Sep}
(6) 
_{Oct}
(2) 
_{Nov}

_{Dec}

2012 
_{Jan}

_{Feb}
(7) 
_{Mar}

_{Apr}
(4) 
_{May}

_{Jun}
(3) 
_{Jul}

_{Aug}

_{Sep}
(1) 
_{Oct}

_{Nov}

_{Dec}
(10) 
2013 
_{Jan}

_{Feb}
(1) 
_{Mar}
(7) 
_{Apr}
(2) 
_{May}

_{Jun}

_{Jul}
(9) 
_{Aug}

_{Sep}

_{Oct}

_{Nov}

_{Dec}

2014 
_{Jan}
(14) 
_{Feb}

_{Mar}
(2) 
_{Apr}

_{May}
(10) 
_{Jun}

_{Jul}

_{Aug}

_{Sep}

_{Oct}

_{Nov}
(3) 
_{Dec}

2015 
_{Jan}

_{Feb}

_{Mar}

_{Apr}

_{May}

_{Jun}

_{Jul}

_{Aug}

_{Sep}

_{Oct}
(12) 
_{Nov}

_{Dec}
(1) 
2016 
_{Jan}

_{Feb}
(1) 
_{Mar}
(1) 
_{Apr}
(1) 
_{May}

_{Jun}
(1) 
_{Jul}

_{Aug}
(1) 
_{Sep}

_{Oct}

_{Nov}

_{Dec}

S  M  T  W  T  F  S 

1
(5) 
2
(9) 
3
(1) 
4
(2) 
5

6
(3) 
7

8

9

10
(1) 
11
(3) 
12

13
(4) 
14

15

16
(2) 
17
(7) 
18
(3) 
19
(5) 
20
(1) 
21

22

23

24
(1) 
25
(7) 
26
(9) 
27
(3) 
28

29
(1) 
30
(10) 
31
(5) 




From: <Thomas_Fields@sc...>  20050517 04:58:56

I will be out of the office starting 17/05/2005 and will not return until 25/05/2005. I will respond to your message when I return. Thomas Fields Technology Group Sony Computer Entertainment Europe ____________________________________ If you require technical support, please use the following methods of contact: Developer Support Hotline No: +44 (0)20 7911 7711 PSP Support: scee_support@... https://psp.scedev.net/ PS2 Support: scee_support@... https://www.ps2pro.com/ PS1 Support: ps1_support@... https://www.ps1pro.com/ ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify postmaster@... This footnote also confirms that this email message has been checked for all known viruses. ********************************************************************** Sony Computer Entertainment Europe 
From: Jonathan Blow <jon@nu...>  20050517 04:54:39

I have some inertia tensor stuff, including a derivation that I think is simpler to understand than the Eberly one, here: http://numbernone.com/blow/inertia/index.html Stan Melax had some stuff up too but I can't seem to find it right now (no obvious link to it from his home page). David Black wrote: > Hi, > >> http://kwon3d.com/theory/jtorque/jtorque.html >> >> But I don`t quite understand how to get the "rate of >> change in angular momentum" as is described in the >> last section. >> > (hope I am not missing something and it is more complicated than it > looks) > > You should probably look at some rigid body physics tutorial for a > description of inertia tensors and skew symetric matrices etc. > > http://www.d6.com/users/checker/dynamics.htm > > Has lots of info, I think his last GDMag article covers working with > inertia tensors etc. But if not there are a number of other places on > there which describe this. > > David > > >  > This SF.Net email is sponsored by Oracle Space Sweepstakes > Want to be the first software developer in space? > Enter now for the Oracle Space Sweepstakes! > http://ads.osdn.com/?ad_id=7412&alloc_id=16344&op=click > _______________________________________________ > GDAlgorithmslist mailing list > GDAlgorithmslist@... > https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist > Archives: > http://sourceforge.net/mailarchive/forum.php?forum_id=6188 > 
From: David Black <dblack@fa...>  20050516 18:29:21

Hi, >http://kwon3d.com/theory/jtorque/jtorque.html > >But I don`t quite understand how to get the "rate of >change in angular momentum" as is described in the >last section. > (hope I am not missing something and it is more complicated than it looks) You should probably look at some rigid body physics tutorial for a description of inertia tensors and skew symetric matrices etc. http://www.d6.com/users/checker/dynamics.htm Has lots of info, I think his last GDMag article covers working with inertia tensors etc. But if not there are a number of other places on there which describe this. David 
From: Sindharta Tanuwijaya <zaraasran@ya...>  20050516 13:01:52

Hi, Thanks to Awen Limbourg who helped me so much the other day, and now I`ve come to the next step, which is finding the torque of each joints, provided that the orientation angles are known for all frames and all joints. I`ve tried to search for some resources in the internet, and one of the resources that I think is good is this: http://kwon3d.com/theory/jtorque/jtorque.html But I don`t quite understand how to get the "rate of change in angular momentum" as is described in the last section. Perhaps someone could help me ? Thank you very much. Sindharta T. __________________________________ Yahoo! Mail Mobile Take Yahoo! Mail with you! Check email on your mobile phone. http://mobile.yahoo.com/learn/mail 
From: Alex Mohr <amohr@cs...>  20050513 20:25:25

>The idea behind the solution discussed in the article is that if we >can somehow minimise our objective function in SO(3) _directly_  >there is no need for the constraints at all. The problem then turns >into one that can be solved by quadratic programming! Just to say it again, for the original problem of finding the 'optimal' rigid transformation between corresponded point clouds, if you're okay with the error metric being a leastsquares thing, you can get away with doing an SVD. Easy to code and stable, robust, etc. I strongly recommend against doing nonlinear optimization for this. Alex 
From: Per Vognsen <Per.Vognsen@ep...>  20050513 18:52:08

> Original Message > From: gdalgorithmslistadmin@... [mailto:gdalgorithms > listadmin@...] On Behalf Of Willem de Boer > Sent: Friday, May 13, 2005 7:30 AM > To: gdalgorithmslist@... > Subject: [Algorithms] Finding optimal transformations >=20 > Everyone, >=20 > In light of a recent discussion on finding a single > Euclidean transformation that "optimally" transforms a > set of points onto another set of the same size, I > recently found a nice little article that some of > the people on this list might find interesting. >=20 > It can be found at: >=20 > http://www.cis.upenn.edu/~cjtaylor/publications/Minimization.pdf >=20 > Consider the problem where we have to find a rotation > matrix that "optimally" rotates a set of points onto > another set. The usual way of going about solving this > is to use some kind of _constrained_ optimisation > algorithm. The constraints are that the matrix be > orthogonal and have determinant 1. SO(n) is a nonsingular affine variety in R^(n^2) so you can use Lagrange multipliers to solve optimization problems on it. The same goes for essentially all the other matrix Lie groups people are interested in. Even slightly funky ones like SE(3) can be handled this way. Recall that SE(3) as a Lie group splits as the semidirect product of R^3 and SO(3). As a differential manifold SE(3) is just the product of R^3 and SO(3), and thus the Jacobian splits in a natural way into the Jacobians of the factors. So optimization on SE(3) is just "simultaneous optimization" on R^3 and SO(3). > The idea behind the solution discussed in the article > is that if we can somehow minimise our objective > function in SO(3) _directly_  there is no need for > the constraints at all. The problem then turns into > one that can be solved by quadratic programming! Quadratic programming is pretty hardcore. Even special cases like convex quadratic programming are darn tricky. Certainly this is a nice geometric way to look at the problem but I think there are probably easier, more efficient algorithms in practice. If you are interested in areas where the theory of Lie groups and algebras can be applied to game development, it is worth looking into "geometric integration". It is the application of Lie algebra techniques to the numerical solution of ODEs on Lie groups. For instance, the dynamics of a rigid body are really the dynamics of a point on SO(3) satisfying the NewtonEuler equations of motion. Usually these are numerically integrated forward in time by regarding a point in SO(3) by its embedding in R^(3x3). If you start with a point on SO(3) and use such a scheme to integrate the point forward in time, you will generally wind up with a point that is no longer on SO(3). The theory of geometric integration instead looks at the corresponding equations of motion on the Lie algebra of SO(3) and "linearizes" them there. The update delta is computed in the Lie algebra and then lifted to the Lie group via the exponential map. The lifted delta is then multiplied onto the point, giving another point of SO(3) since it is closed under multiplication. It turns out that there are integration schemes on the Lie algebra corresponding to RungeKutta, etc. It's all very pretty and some of it is even useful. :) Cheers, Per 
From: PeterPike Sloan <ppsloan@wi...>  20050513 17:46:16

For the problem posted the other methods are superior  they are much simpler than solving a constrained nonlinear optimization problem. The technique presented in the paper is similar to trust region based methods in nonlinear optimization also. It is also possible to solve these (and other problems on these manifolds) using lagrange multipliers and unconstrained nonlinear optimization, which isn't that terrible (still much worse than the techniques in the other thread.) I've done this both on the sphere in R3 and R4 to generalize "weighted averages" for blending points on the sphere or unit quaternions (the weighted average of a set of vectors is the "X" that minimizes the function sum_i(w_i*Distance(P_i,X)^2) in euclidean space (*), replace the distance function with distance over the manifold and you generalize slerp...) The technique only finds a local minimum, which I think might not be a problem for the particular problem, but can be a problem in general. =20 It is an interesting paper though. PeterPike (*) the proof of this is very simple, Distance(A,B)^2 =3D dot(AB,AB), minimize the sum above (wrt X), the second derivative is always positive, so just solve the first derivative for zero. =20 Err(X) =3D sum_i(w_i*sum_j( (P_ijX_j)^2 ) ) The derivative is: dErr/dj =3D sum_i(2*w_i*(X_jP_ij)) Solving equal to zero (each dimension is independent), you get X_j =3D sum_i(w_i*P_ij)/sum_i(w_i) Which is just the weighted average (if the weights sume to 1). Original Message From: gdalgorithmslistadmin@... [mailto:gdalgorithmslistadmin@...] On Behalf Of Willem de Boer Sent: Friday, May 13, 2005 4:30 AM To: gdalgorithmslist@... Subject: [Algorithms] Finding optimal transformations Everyone, In light of a recent discussion on finding a single Euclidean transformation that "optimally" transforms a set of points onto another set of the same size, I recently found a nice little article that some of the people on this list might find interesting. It can be found at: http://www.cis.upenn.edu/~cjtaylor/publications/Minimization.pdf Consider the problem where we have to find a rotation matrix that "optimally" rotates a set of points onto another set. The usual way of going about solving this is to use some kind of _constrained_ optimisation algorithm. The constraints are that the matrix be orthogonal and have determinant 1. The idea behind the solution discussed in the article is that if we can somehow minimise our objective function in SO(3) _directly_  there is no need for the constraints at all. The problem then turns into one that can be solved by quadratic programming! THE COOL THING IS: The cool thing is that this doesn't just work for SO(3) only. The authors give a few examples of where we can translate a constrained optimisation problem into an unconstrained one in SO(3). For example, it is also possible to optimise functions defined on the unit sphere this way! The reason all this works is due to some results in matrix Lie group theory, an introduction to which can be found in the book "Lie Groups, Lie algebras and representations" by Brian C. Hall. It is highly recommended. Cheers,  Willem H. de Boer Homepage: http://www.whdeboer.com=20  This SF.Net email is sponsored by Oracle Space Sweepstakes Want to be the first software developer in space? Enter now for the Oracle Space Sweepstakes! http://ads.osdn.com/?ad_ids93&alloc_id=16281&op=3Dick _______________________________________________ GDAlgorithmslist mailing list GDAlgorithmslist@... https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist Archives: http://sourceforge.net/mailarchive/forum.php?forum_ida88 
From: Willem de Boer <wdeboer@pl...>  20050513 11:30:43

Everyone, In light of a recent discussion on finding a single Euclidean transformation that "optimally" transforms a set of points onto another set of the same size, I=20 recently found a nice little article that some of the people on this list might find interesting. It can be found at: http://www.cis.upenn.edu/~cjtaylor/publications/Minimization.pdf Consider the problem where we have to find a rotation matrix that "optimally" rotates a set of points onto another set. The usual way of going about solving this is to use some kind of _constrained_ optimisation algorithm. The constraints are that the matrix be orthogonal and have determinant 1. The idea behind the solution discussed in the article is that if we can somehow minimise our objective function in SO(3) _directly_  there is no need for the constraints at all. The problem then turns into one that can be solved by quadratic programming! THE COOL THING IS: The cool thing is that this doesn't just work for SO(3) only. The authors give a few examples of where we can translate a constrained optimisation problem into an unconstrained one in SO(3). For example, it is also possible to optimise functions defined on the unit sphere this way! The reason all this works is due to some results in matrix Lie group theory, an introduction to which can be found in the book "Lie Groups, Lie algebras and representations" by Brian C. Hall. It is highly recommended. Cheers,  Willem H. de Boer Homepage: http://www.whdeboer.com=20 
From: Rowan Wyborn <rowan@ir...>  20050511 07:35:07

cool, my eyes just glazed over for that part of the paper... i didnt = realize it was such a simple operator :) Most HDR implementations = already calculate a blurred version of the framebuffer for full screen = glow effects... you could just piggy back on to that, although the paper = seems to imply that the filter width makes a big difference to the final = image quality. > Original Message > From: Matt Pharr [mailto:mattgdalgorithms@...] > Sent: Wednesday, 11 May 2005 1:57 PM > To: gdalgorithmslist@... > Subject: Re: [Algorithms] HDR, Adaptive Luminance, and Exposure >=20 >=20 >=20 > "Rowan Wyborn" <rowan@...> writes: > > I use a simplified verison of their operator, basically=20 > equation (4) in the > > original paper. In this case exposure is controlled by=20 > scaling the Lwhite > > paramater. > > > > I never tried implementing the full operator because=20 > applying the gaussian > > convolution to the entire scene seems like overkill=20 > (especially given how > > slow most hardware is at operating on FP textures). Certainly the > > simplified operator is giving me really good results (much=20 > better than a > > pure linear tonemap), and it is easy for artists to control. >=20 > All that Gaussian stuff in their paper is just to find the=20 > local contrast > in the image, which is defined at each pixel (x,y) as: >=20 > lc(s,x,y) =3D (B(s,x,y)  B(2s,x,y)) / B(s,x,y), >=20 > where s is a filter width and B is a blurred version of the image with > filter width s. They used a Gaussian for computing the=20 > blurred version of > the image, though I don't think that there's a fundamental=20 > need that it be > a Gaussian. >=20 > In software, I've gotten decent results using straightup MIPmaps to > compute those blurred images. This approach would be an=20 > option for GPUs > that support blending of floatingpoint imagesI'd guess it'd be > reasonably fast! >=20 > matt > =20 > Matt Pharr matt@... =20 > <URL:http://graphics.stanford.edu/~mmp>; > = =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > =3D=3D=3D=3D=3D=3D=3D=3D=3D > In a cruel and evil world, being cynical can allow you to get some > entertainment out of it. Daniel Waters >=20 >=20 >  > This SF.Net email is sponsored by Oracle Space Sweepstakes > Want to be the first software developer in space? > Enter now for the Oracle Space Sweepstakes! > http://ads.osdn.com/?ad_id=3D7393&alloc_id=3D16281&op=3Dclick > _______________________________________________ > GDAlgorithmslist mailing list > GDAlgorithmslist@... > https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist > Archives: > http://sourceforge.net/mailarchive/forum.php?forum_id=3D6188 >=20 
From: Matt Pharr <mattgdalgorithms@ph...>  20050511 04:00:51

"Rowan Wyborn" <rowan@...> writes: > I use a simplified verison of their operator, basically equation (4) in the > original paper. In this case exposure is controlled by scaling the Lwhite > paramater. > > I never tried implementing the full operator because applying the gaussian > convolution to the entire scene seems like overkill (especially given how > slow most hardware is at operating on FP textures). Certainly the > simplified operator is giving me really good results (much better than a > pure linear tonemap), and it is easy for artists to control. All that Gaussian stuff in their paper is just to find the local contrast in the image, which is defined at each pixel (x,y) as: lc(s,x,y) = (B(s,x,y)  B(2s,x,y)) / B(s,x,y), where s is a filter width and B is a blurred version of the image with filter width s. They used a Gaussian for computing the blurred version of the image, though I don't think that there's a fundamental need that it be a Gaussian. In software, I've gotten decent results using straightup MIPmaps to compute those blurred images. This approach would be an option for GPUs that support blending of floatingpoint imagesI'd guess it'd be reasonably fast! matt  Matt Pharr matt@... <URL:http://graphics.stanford.edu/~mmp>; ======================================================================= In a cruel and evil world, being cynical can allow you to get some entertainment out of it. Daniel Waters 
From: Rowan Wyborn <rowan@ir...>  20050511 00:18:11

I use a simplified verison of their operator, basically equation (4) in = the original paper. In this case exposure is controlled by scaling the = Lwhite paramater. I never tried implementing the full operator because applying the = gaussian convolution to the entire scene seems like overkill (especially = given how slow most hardware is at operating on FP textures). Certainly = the simplified operator is giving me really good results (much better = than a pure linear tonemap), and it is easy for artists to control. rowan > Original Message > From: Brad_Byrd@... > [mailto:Brad_Byrd@...] > Sent: Wednesday, 11 May 2005 6:54 AM > To: gdalgorithmslist@... > Cc: Andy_Styles@... > Subject: [Algorithms] HDR, Adaptive Luminance, and Exposure >=20 >=20 > So I've implemented the Reinhard et. al. method of doing tone mapping=20 > based on the measured average luminance of the scene. (See the=20 > "HDRLighting" sample in the latest D3D SDK as well as=20 > http://www.cs.utah.edu/~reinhard/cdrom/tonemap.pdf for the=20 > original paper=20 > and equations.) The results are quite nice. However, nowhere in the=20 > paper or in that sample is there any kind of "exposure"=20 > setting. Many=20 > other HDR samples have a usertweakable exposure setting. My=20 > question is,=20 > how does a manual exposure setting factor into this method? =20 > I'm not sure=20 > exactly where to multipy in an exposure value in the pixel=20 > shader to give=20 > the right effect. Or, does the nature of this method=20 > preclude the use of=20 > an exposure setting because of the way that the adaptive luminance=20 > automatically adjusts the brightness of the final image? >=20 > Also, on the same topic, while the results look nice it looks=20 > like there=20 > may be a difficult road ahead for artists that need to dial=20 > in correct=20 > middlegray, exposure, brightness threshold, brightness=20 > offset, luminance=20 > adapt rate, and gaussian blur radius values to get a scene to=20 > look good.=20 > There are lots and lots of unintuitive tweakable values that=20 > need to be=20 > setup. Do any of you have experience in dealing with this=20 > issue? Maybe a=20 > less physically correct but simpler tonemapping operator=20 > that doesn't use=20 > adaptive luminance would be better in the long run as there=20 > are much fewer=20 > tweakables? >=20 > Thoughts? >=20 > Thanks, >=20 > Brad... >=20 >=20 >=20 >  > This SF.Net email is sponsored by Oracle Space Sweepstakes > Want to be the first software developer in space? > Enter now for the Oracle Space Sweepstakes! > http://ads.osdn.com/?ad_id=3D7393&alloc_id=3D16281&op=3Dclick > _______________________________________________ > GDAlgorithmslist mailing list > GDAlgorithmslist@... > https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist > Archives: > http://sourceforge.net/mailarchive/forum.php?forum_id=3D6188 >=20 
From: <Brad_Byrd@pl...>  20050510 20:52:07

So I've implemented the Reinhard et. al. method of doing tone mapping based on the measured average luminance of the scene. (See the "HDRLighting" sample in the latest D3D SDK as well as http://www.cs.utah.edu/~reinhard/cdrom/tonemap.pdf for the original paper and equations.) The results are quite nice. However, nowhere in the paper or in that sample is there any kind of "exposure" setting. Many other HDR samples have a usertweakable exposure setting. My question is, how does a manual exposure setting factor into this method? I'm not sure exactly where to multipy in an exposure value in the pixel shader to give the right effect. Or, does the nature of this method preclude the use of an exposure setting because of the way that the adaptive luminance automatically adjusts the brightness of the final image? Also, on the same topic, while the results look nice it looks like there may be a difficult road ahead for artists that need to dial in correct middlegray, exposure, brightness threshold, brightness offset, luminance adapt rate, and gaussian blur radius values to get a scene to look good. There are lots and lots of unintuitive tweakable values that need to be setup. Do any of you have experience in dealing with this issue? Maybe a less physically correct but simpler tonemapping operator that doesn't use adaptive luminance would be better in the long run as there are much fewer tweakables? Thoughts? Thanks, Brad... 
From: Rodrigo H. <phamtompain@ya...>  20050506 10:48:03

Hi, Black and White 2 (from Lionhead studios, Peter Molyneux) had realtime terrain "shaping" (not sure if such a word exists =) backed by a physical engine. He presented the game at GDC (you can watch his presentation through GDCTV at http://www.pqhp.com/cmp/gdctv/) and maybe it may interest you, give you tips of what you should look for. Rodrigo Greselle Hartmann Mensagem original De: gdalgorithmslistadmin@... [mailto:gdalgorithmslistadmin@...] Em nome de Ryan De Boer Enviada em: sextafeira, 6 de maio de 2005 00:58 Para: gdalgorithmslist@... Assunto: [Algorithms] large terrain with lots of dynamic changes We want a very large terrain with lots of dynamic changes that can happen locally in different areas. We are worried about, for example having one big vertex buffer and having to send lots of data to the graphics card every frame (which doesn't scale well). We also want an algorithm that allows it to be easy to modify the terrain's visual quality by an artist. Maybe some of you with experience deforming terrain dynamically with explosions etc can shed some light on some algorithms that might be good to look at. A lot of the ideas from static terrain algorithms may not work well for dynamic terrain. I know one answer will be ROAM, but are there any others? Any papers to read would be helpful. Ryan De Boer This message contains confidential information and is intended only for the individual(s) named. If you are not the named addressee you should not disseminate, distribute or copy this email. Please immediately delete it and all copies of it from your system, destroy any hard copies of it, and notify the sender. Email transmission cannot be guaranteed to be secure or errorfree as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or contain viruses. To the maximum extent permitted by law, Immersive Technologies Pty. Ltd. does not accept liability for any errors or omissions in the contents of this message which arise as a result of email transmission.  This SF.Net email is sponsored by: NEC IT Guy Games. Get your fingers limbered up and give it your best shot. 4 great events, 4 opportunities to win big! Highest score wins.NEC IT Guy Games. Play to win an NEC 61 plasma display. Visit http://www.necitguy.com/?r _______________________________________________ GDAlgorithmslist mailing list GDAlgorithmslist@... https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist Archives: http://sourceforge.net/mailarchive/forum.php?forum_ida88 
From: Alexey Kryazhev <alexey.kryazhev@co...>  20050506 10:45:59

I presume the landscape is not changed every frame. What if instead of a big dynamic buffer you use a few static vertex buffers for your terrain chunks and update them only when it's necessary. Static buffers with infrequent updates worked quite well for me at least. Alexey Original Message From: gdalgorithmslistadmin@... [mailto:gdalgorithmslistadmin@...]On Behalf Of Ryan De Boer Sent: 06 May 2005 04:58 To: gdalgorithmslist@... Subject: [Algorithms] large terrain with lots of dynamic changes We want a very large terrain with lots of dynamic changes that can happen locally in different areas. We are worried about, for example having one big vertex buffer and having to send lots of data to the graphics card every frame (which doesn't scale well). We also want an algorithm that allows it to be easy to modify the terrain's visual quality by an artist. Maybe some of you with experience deforming terrain dynamically with explosions etc can shed some light on some algorithms that might be good to look at. A lot of the ideas from static terrain algorithms may not work well for dynamic terrain. I know one answer will be ROAM, but are there any others? Any papers to read would be helpful. Ryan De Boer This message contains confidential information and is intended only for the individual(s) named. If you are not the named addressee you should not disseminate, distribute or copy this email. Please immediately delete it and all copies of it from your system, destroy any hard copies of it, and notify the sender. Email transmission cannot be guaranteed to be secure or errorfree as information could be intercepted, corrupted, lost, destroyed, arrive late or incomplete, or contain viruses. To the maximum extent permitted by law, Immersive Technologies Pty. Ltd. does not accept liability for any errors or omissions in the contents of this message which arise as a result of email transmission.  This SF.Net email is sponsored by: NEC IT Guy Games. Get your fingers limbered up and give it your best shot. 4 great events, 4 opportunities to win big! Highest score wins.NEC IT Guy Games. Play to win an NEC 61 plasma display. Visit http://www.necitguy.com/?r _______________________________________________ GDAlgorithmslist mailing list GDAlgorithmslist@... https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist Archives: http://sourceforge.net/mailarchive/forum.php?forum_ida88 ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the system manager. This footnote also confirms that this email message has been swept by MIMEsweeper for the presence of computer viruses. ********************************************************************** 
From: Ryan De Boer <RDeBoer@im...>  20050506 03:58:19

We want a very large terrain with lots of dynamic changes that can happen locally in different areas. We are worried about, for example having one big vertex buffer and having to send lots of data to the graphics card every frame (which doesn't scale well). We also want an algorithm that allows it to be easy to modify the terrain's visual quality by an artist. Maybe some of you with experience deforming terrain dynamically with explosions etc can shed some light on some algorithms that might be good to look at. A lot of the ideas from static terrain algorithms may not work well for dynamic terrain. I know one answer will be ROAM, but are there any others? Any papers to read would be helpful. Ryan De Boer This message contains confidential information and is intended only for = the individual(s) named. If you are not the named addressee you should not = disseminate, distribute or copy this email. Please immediately delete it = and all copies of it from your system, destroy any hard copies of it, and = notify the sender. Email transmission cannot be guaranteed to be secure = or errorfree as information could be intercepted, corrupted, lost, = destroyed, arrive late or incomplete, or contain viruses. To the maximum = extent permitted by law, Immersive Technologies Pty. Ltd. does not accept = liability for any errors or omissions in the contents of this message = which arise as a result of email transmission.= 
From: Mark Duchaineau <duchaine@ll...>  20050504 17:01:13

Hi Brian, Yes, a0 is the quadtree ancestor (so the structure includes conventional quadtrees as one of its native capabilities), a1 and a3 are the right and left parents (that each overlap half of the diamond, thus defining conventional "triangle bintree" structures). The a2 ancestor is kind of an oddball. It can be a grandparent, or at any coarser level. Usually the a2 ancestor is only used to grab that corner vertex, and is not used in its role as an edge or face. We review the BDAM/PBDAM work in the papers (briefly). I like their work a lot. The main issues are with dynamic geometry, filtering, textures and compression, all of which I think work out better using diamond tiles. Their use of high quality unstructured (TIN) simplification can produce lower geometric error for a given triangle count. There is some cost to this for preprocessing, and because you have a unique index array per triangle patch (with regular tiles you can just reuse a single index array for all patches). Also, the lowpass geometry filtering of regular tiles is very high quality using well known image processing methods, whereas the TIN methods are much trickier to use to avoid unintentional artifacts (think lighting and interpolation issues). My argument is that given the huge triangle counts we can now achieve on highend (soon to be lowend) PC graphics cards, the process should be though of as doing high quality filtering for both textures and geometry. BTW, a somewhat shortened version of the paper will appear in TVCG soon. The best version to date is at http://graphics.cs.ucdavis.edu/~duchaine/roam_lecture20041110/hwa_tvcg04_revised.pdf I am working on getting some demo code and implementation notes out on the ROAM homepage. Cheers, Mark D. Brian Hook wrote: > Thread necromancy at work! > > I read Hwa's paper, but I'm a little confused with some of the stuff > in there to do with the diamond topology. In Fig 4 of the revised > IEEE paper, the ancestors are labeled a0 to a3. a0 is the ancestor > quad tree, a1/a2 are the immediate parents, but what is a2? It's not > clear from the diagram what that's supposed to be. > > >>You should also check out the BDAM papers by Cignoni and his group. > > > How would you contrast the PBDAM stuff with Hwa's work? > > Brian > > > > >  > This SF.Net email is sponsored by: NEC IT Guy Games. > Get your fingers limbered up and give it your best shot. 4 great events, 4 > opportunities to win big! Highest score wins.NEC IT Guy Games. Play to > win an NEC 61 plasma display. Visit http://www.necitguy.com/?r > _______________________________________________ > GDAlgorithmslist mailing list > GDAlgorithmslist@... > https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist > Archives: > http://sourceforge.net/mailarchive/forum.php?forum_ida88 > 
From: <Paul_Firth@sc...>  20050504 08:48:26

gdalgorithmslistadmin@... wrote on 30/04/2005 18:22:55: > hi, > your tutorials are helpful, however.. it seems that > you always end up with a vertexedge contact (i.e a > single contact point). You're still talking about 2d, right? > do you have any suggestions for how to handle > edgeedge cases, where there are two contact points? > (i.e when there are two colinear edges on the convex > hull) There shouldn't be a problem here if you treat your MD edges as infinite lines (obviously this only works for the penetration case, not the positive distance case). > i've tried "fixing" this in the MDconstruction phase > (detecting parallel edges there and making a single MD > edge instead of two colinear ones), and in the > calculatepenetration phase, however none of the stuff > i've tried thus far has been as simple, nice, and > wellbehaved as the vertexedge stuff (which works > every time). I think i've understood the question, but feel free to tell me otherwise! ;) Cheers, Paul. ********************************************************************** This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify postmaster@... This footnote also confirms that this email message has been checked for all known viruses. ********************************************************************** Sony Computer Entertainment Europe 
From: Gino van den Bergen <gvandenbergen@pl...>  20050503 09:46:16

The SAT will catch all intersecting cases but it may also return false = positives for some degenerate cases. For instance, if a pair of edges = from either triangle is close to being collinear, the cross product of = the two edge vectors can be quite noisy and may not be able to separate = the triangles. =20 Mueller's approach does not detect intersecting coplanar triangles, but = I find it hard to believe that this accounts for 17% fewer collisions, = unless you are testing the mesh against itself. =20 =20 Have you looked at Guigue's test = http://home.arcor.de/philippe.guigue/triangle_triangle_intersection.htm = ?=20 =20 Another option is to use GJK from SOLID 3.5 which can be downloaded from = http://www.dtecta.com. Do not forget to use the robust version by = uncommenting //#define JOHNSON_ROBUST in the file DT_GJK.h. Also, you = might want to try to use doubles for internal computations (define = DUSE_DOUBLES as preprocessor option.). =20 Please keep us posted on your findings. =20 Best, =20 Gino =20 _____ =20 From: gdalgorithmslistadmin@... = [mailto:gdalgorithmslistadmin@...] On Behalf Of = Gilvan at luke Sent: Monday, May 02, 2005 6:28 PM To: gdalgorithmslist@... Subject: [Algorithms] TriTri Intersection: SATbased, M=F6ller or = both? =09 =09 I did some tests between two algorithms for collision detection between = a pair of stationary triangles:=20  Th=F6mas Mollher's one, from his paper "A Fast TriangleTriangle = Intersection Test"; (code from Opcode, welldesigned for every case  a bit complicated = )  SATbased (code from MagikSoftware, "boosted" to deal better with the = coplanar case  easy understanding ) =20 First of all, I think is reasonable to choose the algorithm that is = best suited (robust, fast) for CLOSE pairs, because my prebuilt BV tree = will reject the distant ones. =20 My tests pointed that the SATbased algorithm is faster... but it = detects 17% more collisions than the other one. =20 Did anybody do any related tests and share the results ? : ) =20 By the way, can anybody show me an algorithm that is VERY robust to = compare the results of other algoritms with ? =20 =20 Gilvan Maia 
From: <Simon.Fenney@po...>  20050502 21:25:52

If you're after the principal eigenvalue/vector (assuming there is a unique= one) you could always use the "repeatedly square the matrix" approach. I u= se this in a texture compressor (with a 4x4 (i.e. RBGA) matrix) and it seem= s converge quite rapidly.=0D=0A=20=0D=0ASimon=0D=0A=20=0D=0A=0D=0AOrig= inal Message=0D=0AFrom: gdalgorithmslistadmin@... = [mailto:gdalgorithmslistadmin@...]On Behalf Of Bill Bax= ter=0D=0ASent: 02 May 2005 10:29=0D=0ATo: gdalgorithmslist@...= ge.net=0D=0ASubject: Re: [Algorithms] Best fit rigid body rotation for a se= t of points=0D=0A=0D=0A=0D=0AThanks Alex, glad to see I wasn't wrong about = you "mocap dudes" being on top of this stuff. :)=0D=0A=0D=0AThat's a great= reference. Compares 4 different methods in detail, including the Horn's m= ethod PeterPike mentioned. Now that I look more closely, the tricky part = of this Horn's method is that it requires finding the most positive root of= a quartic equation (to find the most positive eigenvalue of a 4x4 matrix),= and Horn doesn't really offer any suggestions for doing that other than to= remark that closed form solutions exist. However, if I read it right it l= ooks like you know that the eigenvalues of this matrix are all real and all= have magnitude <1. So given that, it seems like Newton's method starting = from lambda=3D1 would converge pretty quickly to the most positive root.=0D= =0A=0D=0AAnyway, according the ref you linked to, Horn's orthonormal matrix= based method is the fastest for smallish data sets, which is what I'm afte= r.=0D=0A=0D=0ABill=0D=0A=0D=0A=0D=0AOn 5/2/05, Alex Mohr < amohr@...= edu> wrote:=20=0D=0A=0D=0AJoining late, so I apologize if this is repeating= but this may be=0D=0Auseful:=0D=0A=0D=0Ahttp://www.cs.duke.edu/researchers= /artificial_intelligence/temp/eggert_rigid_body_transformations.pdf <http:= //www.cs.duke.edu/researchers/artificial_intelligence/temp/eggert_rigid_bod= y_transformations.pdf>=20=0D=0A=0D=0AGenerally, I believe you should be abl= e to get away with nothing more=0D=0Aannoying than doing an SVD.=0D=0A=0D=0A= Alex=0D=0A=0D=0A=0D=0A>Great, that looks like exactly what I was after. I f= igured vision folks or=0D=0A>mocap dudes among others must have run into th= e same problem, but the search=20=0D=0A>terms I tried were all turning up b= ubkis.=0D=0A>=0D=0A>With the info in the paper you linked to I was able to = find the Horn paper=0D=0A>here:=0D=0A> <http://people.csail.mit.edu/u/b/bk= ph/public_html/papers/AbsoluteOPT.pdf> http://people.csail.mit.edu/u/b/bkp= h/public_html/papers/AbsoluteOPT.pdf=0D=0A>=0D=0A>And I found this useful = overview of a variety of similar techniques.=0D=0A> <http://www.cs.duke.ed= u/~jeffp/triseminar/absoluteorientation.pdf> http://www.cs.duke.edu/~jeffp= /triseminar/absoluteorientation.pdf=0D=0A>=0D=0A>Thanks!=0D=0A>bb=0D=0A>=0D= =0A>On 5/1/05, PeterPike Sloan < ppsloan@... <mailto:pp= sloan@...> > wrote:=0D=0A>>=0D=0A>> Google for "procruste= s method", this can be done without resorting to=0D=0A>> nonlinear optimiz= ation... I think in the vision community they call it=0D=0A>> "horns method= ", or something like that (some late 80's vision paper.) Here's=20=0D=0A>> = a paper that deals with it:=0D=0A>>=0D=0A>> http://www.photogrammetry.ethz.= ch/general/persons/devrim/LS3D_04WS_Dresden.pdf <http://www.photogrammetry= =2Eethz.ch/general/persons/devrim/LS3D_04WS_Dresden.pdf>=20=0D=0A>> Pete= rPike=0D=0A>>=0D=0A>> =0D=0A>> *From:* gdal= gorithmslistadmin@... [mailto:=20=0D=0A>> gdalgorithms= listadmin@...] *On Behalf Of *Bill Baxter=0D=0A>> *Sent:= * Saturday, April 30, 2005 4:47 PM=0D=0A>> *To:* gdalgorithmslist@...= urceforge.net=0D=0A>> *Subject:* [Algorithms] Best fit rigid body rotation = for a set of points=0D=0A>>=0D=0A>> Does anyone know an algorithm for deter= mining the rigid body=20=0D=0A>> transformation that best fits the motion o= f a set of N points=3F=0D=0A>>=0D=0A>> I.e. given a set of initial points p= 0 p1 ... pN, and their locations after=0D=0A>> some motion, q0 q1 ... qN, I= want to find the rigid body transformation T(x)=20=0D=0A>> (rotation and t= ranslation only) that minimizes:=0D=0A>> Sum[i=3D0...N] (  T(pi)  qi ^= 2 )=0D=0A>>=0D=0A>> The only thing I can think of is to treat it as a gener= ic nonlinear=0D=0A>> optimization problem, but it just seems like there sho= uld be something more=20=0D=0A>> efficient than that.=0D=0A>>=0D=0A>>=0D=0A= >> Bill=0D=0A>>=0D=0A>>=0D=0A=0D=0A= =0D=0AThis SF.Net email is sponsored by: NEC IT Guy Games.= =20=0D=0AGet your fingers limbered up and give it your best shot. 4 great e= vents, 4=0D=0Aopportunities to win big! Highest score wins.NEC IT Guy Games= =2E Play to=0D=0Awin an NEC 61 plasma display. Visit http://www.necitguy.co= m/=3Fr=3D20=0D=0A_______________________________________________=0D=0AGDAlg= orithmslist mailing list=0D=0AGDAlgorithmslist@...=20=0D= =0Ahttps://lists.sourceforge.net/lists/listinfo/gdalgorithmslist=0D=0AArch= ives:=0D=0Ahttp://sourceforge.net/mailarchive/forum.php=3Fforum_id=3D6188=0D= =0A=0D=0A=0D=0A=0D=0A=0D=0A******************=0D=0AThis email has been sen= t from Imagination Technologies Limited.=0D=0APowerVR, Metagence, Ensigma a= nd PURE Digital are divisions=0D=0Aof Imagination Technologies Limited.=0D=0A=0D= =0AThe information contained in this email, including any attachment,=0D=0A= is confidential and may be legally privileged. It is intended solely=0D=0A= for the addressee(s) and access to this email by anyone else is=0D=0Aunaut= horised. If you are not the intended recipient, any disclosure,=0D=0Acopyi= ng or distribution or use of the information contained in this=20=0D=0Aema= il, is prohibited and may be unlawful. If you have received this=0D=0Aema= il in error, please notify the sender by return email and then=0D=0Adelete= it from your system.=0D=0A=0D=0AInternet communications cannot be guarante= ed to be secure,=0D=0Aerror or virusfree. The sender does not accept liab= ility for any errors=0D=0Aor omissions which arise as a result.=0D=0A=0D=0A= Any views expressed in this message are those of the author, except=0D=0Awh= ere the author specifies and, with authority, states them to be the=0D=0Avi= ews of Imagination Technologies Limited.=0D=0A 
From: Gilvan Maia <gilvan@vd...>  20050502 18:47:12

>> By the way, can anybody show me an algorithm that is VERY robust to >> compare the results of other algoritms with ? > > I would personally go the other way and write a program to methodically > generate all of the ikky special cases (and also the cases within an > epsilon of the ikky ones), knowing what results to expect in advance. Yep! This is what I am currently doing.... but my friends around here told me that I was cheating!!! : P So I'm using some wellknown models (torus, sphere and a character) and randomizing orientations accordingly to Ken Shoemake's gem "Uniform Random Rotations". Currently, I am thinking about implementing some simpler algorithm in a bigger precision model and precompute the outputs... Can anyone show me his opinion, a paper or something else about it? Thanks, Gilvan 
From: Stephen J Baker <sjbaker@li...>  20050502 18:39:39

Gilvan at luke wrote: > By the way, can anybody show me an algorithm that is VERY robust to > compare the results of other algoritms with ? (oops hit send too soon!) ...you should also test that (for example) if you have a whole mesh of triangles  and intersect that with a triangle that exactly lines up with an internal edge of the mesh  that the triangle does indeed intersect at least (and preferably, at most) one triangle of that mesh. There are algorithms that allow a triangle to slip right through in such a situation.  The second law of Frisbee throwing states: "Never precede any maneuver by a comment more predictive than "Watch this!"...it turns out that this also applies to writing Fragment Shaders.  Steve Baker (817)6192657 (Vox/VoxMail) L3Com/Link Simulation & Training (817)6192466 (Fax) Work: sjbaker@... http://www.link.com Home: sjbaker1@... http://www.sjbaker.org 
From: Stephen J Baker <sjbaker@li...>  20050502 18:32:23

Gilvan at luke wrote: > By the way, can anybody show me an algorithm that is VERY robust to > compare the results of other algoritms with ? I would personally go the other way and write a program to methodically generate all of the ikky special cases (and also the cases within an epsilon of the ikky ones), knowing what results to expect in advance.  The second law of Frisbee throwing states: "Never precede any maneuver by a comment more predictive than "Watch this!"...it turns out that this also applies to writing Fragment Shaders.  Steve Baker (817)6192657 (Vox/VoxMail) L3Com/Link Simulation & Training (817)6192466 (Fax) Work: sjbaker@... http://www.link.com Home: sjbaker1@... http://www.sjbaker.org 
From: Gilvan at luke <gilvan@vd...>  20050502 16:16:32

I did some tests between two algorithms for collision detection between = a pair of stationary triangles:=20  Th=F6mas Mollher's one, from his paper "A Fast TriangleTriangle = Intersection Test"; (code from Opcode, welldesigned for every case  a bit complicated = )  SATbased (code from MagikSoftware, "boosted" to deal better with the coplanar = case  easy understanding ) First of all, I think is reasonable to choose the algorithm that is best = suited (robust, fast) for CLOSE pairs, because my prebuilt BV tree will = reject the distant ones. My tests pointed that the SATbased algorithm is faster... but it = detects 17% more collisions than the other one. Did anybody do any related tests and share the results ? : ) By the way, can anybody show me an algorithm that is VERY robust to = compare the results of other algoritms with ? Gilvan Maia 
From: Alex Mohr <amohr@cs...>  20050502 16:13:17

>That's a great reference. Compares 4 different methods in detail, including >the Horn's method PeterPike mentioned. Now that I look more closely, the >tricky part of this Horn's method is that it requires finding the most >positive root of a quartic equation (to find the most positive eigenvalue of >a 4x4 matrix), and Horn doesn't really offer any suggestions for doing that >other than to remark that closed form solutions exist. However, if I read it >right it looks like you know that the eigenvalues of this matrix are all >real and all have magnitude <1. So given that, it seems like Newton's method >starting from lambda=1 would converge pretty quickly to the most positive >root. > >Anyway, according the ref you linked to, Horn's orthonormal matrix based >method is the fastest for smallish data sets, which is what I'm after. Yeah, there's a closed form for quartics, but it's a few pages of annoyance. I'd recommend against doing something like Newton's method. If you have smallish datasets, why not try doing the SVD thing? It should be plenty fast and very stable, generally wellbehaved, and also easy to code up. Alex 
From: PeterPike Sloan <ppsloan@wi...>  20050502 14:52:58

You don't need to use newtons method  google for "iterative eigenvectors", the power method/inverse power method are techniques to do this  the power method pulls off the largest eigenvector of a matrix, inverse power can pull of the eigenvector corresponding to the eigenvalue closest to a given value. =20 These are the "conventional" techniques referred to in the paper. One sided jacobi is a simple iterative algorithm that can be used to compute the SVD as well. =20 PeterPike =20 (The technique Horn refers to to solve for the roots of a quartic is due to ferraria: =20 http://www.sosmath.com/algebra/factor/fac12/fac12.html ) =20 ________________________________ From: gdalgorithmslistadmin@... [mailto:gdalgorithmslistadmin@...] On Behalf Of Bill Baxter Sent: Monday, May 02, 2005 2:29 AM To: gdalgorithmslist@... Subject: Re: [Algorithms] Best fit rigid body rotation for a set of points =09 =09 Thanks Alex, glad to see I wasn't wrong about you "mocap dudes" being on top of this stuff. :) =09 That's a great reference. Compares 4 different methods in detail, including the Horn's method PeterPike mentioned. Now that I look more closely, the tricky part of this Horn's method is that it requires finding the most positive root of a quartic equation (to find the most positive eigenvalue of a 4x4 matrix), and Horn doesn't really offer any suggestions for doing that other than to remark that closed form solutions exist. However, if I read it right it looks like you know that the eigenvalues of this matrix are all real and all have magnitude <1. So given that, it seems like Newton's method starting from lambda=3D1 would converge pretty quickly to the most positive root. =09 Anyway, according the ref you linked to, Horn's orthonormal matrix based method is the fastest for smallish data sets, which is what I'm after. =09 Bill =09 =09 On 5/2/05, Alex Mohr <amohr@...> wrote:=20 Joining late, so I apologize if this is repeating but this may be useful: =09 =09 http://www.cs.duke.edu/researchers/artificial_intelligence/temp/eggert_r igid_body_transformations.pdf=20 =09 Generally, I believe you should be able to get away with nothing more annoying than doing an SVD. =09 Alex =09 =09 >Great, that looks like exactly what I was after. I figured vision folks or >mocap dudes among others must have run into the same problem, but the search=20 >terms I tried were all turning up bubkis. > >With the info in the paper you linked to I was able to find the Horn paper >here: > http://people.csail.mit.edu/u/b/bkph/public_html/papers/AbsoluteOPT.pdf <http://people.csail.mit.edu/u/b/bkph/public_html/papers/AbsoluteOPT.pd f>=20 > >And I found this useful overview of a variety of similar techniques. > http://www.cs.duke.edu/~jeffp/triseminar/absoluteorientation.pdf <http://www.cs.duke.edu/~jeffp/triseminar/absoluteorientation.pdf>=20 > >Thanks! >bb > >On 5/1/05, PeterPike Sloan <ppsloan@... > wrote: >> >> Google for "procrustes method", this can be done without resorting to >> nonlinear optimization... I think in the vision community they call it >> "horns method", or something like that (some late 80's vision paper.) Here's=20 >> a paper that deals with it: >> >> http://www.photogrammetry.ethz.ch/general/persons/devrim/LS3D_04WS_Dresd en.pdf=20 >> PeterPike >> >>  >> *From:* gdalgorithmslistadmin@... [mailto:=20 >> gdalgorithmslistadmin@...] *On Behalf Of *Bill Baxter >> *Sent:* Saturday, April 30, 2005 4:47 PM >> *To:* gdalgorithmslist@... >> *Subject:* [Algorithms] Best fit rigid body rotation for a set of points >> >> Does anyone know an algorithm for determining the rigid body=20 >> transformation that best fits the motion of a set of N points? >> >> I.e. given a set of initial points p0 p1 ... pN, and their locations after >> some motion, q0 q1 ... qN, I want to find the rigid body transformation T(x)=20 >> (rotation and translation only) that minimizes: >> Sum[i=3D0...N] (  T(pi)  qi ^2 ) >> >> The only thing I can think of is to treat it as a generic nonlinear >> optimization problem, but it just seems like there should be something more=20 >> efficient than that. >> >> >> Bill >> >> =09  This SF.Net email is sponsored by: NEC IT Guy Games.=20 Get your fingers limbered up and give it your best shot. 4 great events, 4 opportunities to win big! Highest score wins.NEC IT Guy Games. Play to win an NEC 61 plasma display. Visit http://www.necitguy.com/?r=3D20 _______________________________________________ GDAlgorithmslist mailing list GDAlgorithmslist@...=20 =09 https://lists.sourceforge.net/lists/listinfo/gdalgorithmslist Archives: =09 http://sourceforge.net/mailarchive/forum.php?forum_id=3D6188 =09 