[ptx] Help on SVD decompoistion code.

[Xianyong Fang]

Hi Alexandre,

  I appreciate your kindly help.
  Actually I have integrated one svd decomposition into my code. In addition, you have suggested me to check the homography computation in the http://www.cs.unc.edu/~blloyd/comp290-089/fmatrix/. I find one strange phenomena, that is, my current svd code generate a different V(eigen vector matrix). Sometimes(not always!) the V of mine is negative to the V generated in Matlab(that is they have the same absolute value, but all the elements are negative to the other ones. Of course at these times the U is different from that geneated in Matlab). So the final F(the fundamental matrix) of my program is different from that generated in Matlab. I think it is very strange. 

  Of course I don't check the principle of SVD decomposition.

  Have you ever encountered such phenomena? 
	

Xianyong

========================

>BTW you can also use my old highly optimized c++ matrix class :
>http://www.le-geo.com/computer/matrix/index.html
>...
>Unfortunately, when I wrote it, I relied on the Intel MKL which was free.
>Anyway. It's a great matrice implementation (with SVD, QR, etc).
>
>Alexandre
>
>> -----Message d'origine-----
>> De : ptx-bounces at email-lists.org 
>> [mailto:ptx-bounces at email-lists.org] De la part de Xianyong Fang
>> Envoy?: jeudi 8 avril 2004 15:01
>> ?: ptx at email-lists.org
>> Objet : Re: Re: [ptx] Help on SVD decompoistion code.
>> 
>> 
>> Hi Rafal,
>> 
>>   Wonderful. It is really very useful. Thanks very much, :)
>> 
>> 	
>> 
>> Xianyong
>> 
>> ========================
>> 
>> >On Thu, 8 Apr 2004, alexandre jenny wrote:
>> >
>> >> For any numerical algorithm, there's only one thing to know : 
>> >> Numerical Recipies : http://www.nr.com/ You can get the full book 
>> >> online and the svd decomposition is inside "in c or fortran code".
>> >> (the c++ version isn't free but doesn't worth it).
>> >
>> >Unfortunately, the code is not free. You can't distribute 
>> it. I'm not 
>> >sure
>> >if you can use it for your own if you don't buy the book, either. 
>> >
>> >On the other hand www.netlib.org is full of free (as beer and as in
>> >speech) programs.  SVD is a part of LAPACK which is the 
>> standard set of 
>> >routines for linear algebra applications.  Individual routine with 
>> >dependencies in Fortran is here
>> >
>> >http://www.netlib.org/cgi-bin/netlibfiles.pl?filename=/lapack
>/double/dg
>>esvd.f
>>
>>There is a C interface to lapack (http://www.netlib.org/clapack/). 
>>There is also a C++ interface, http://math.nist.gov/lapack++/ which 
>>according to the page is superseded by the Template Numerical Toolkit 
>>(TNT) so you may also look here http://math.nist.gov/tnt/
>>
>>LAPACK is used in many scientific programs and it's well optimized and
>>documented. 
>>
>>Regards,
>>
>>Rafal
>>
>>
>>
>
>= = = = = = = = = = = = = = = = = = = =

= = = = = = = = = = = = = = = = = = = =