[ptx] beta version available for pano12.dll with improved optimizer

Rik Littlefield rj.littlefield at computer.org
Mon May 17 16:46:04 BST 2004


Terje,

I think your basic problem was a glitch in control point definitions.  
Your Line #4 in lens-7.1mm.ptp looks to include two separate lines -- 
like you accidentally forgot to do a New Line when entering the points.  
I deleted the spare points from Line #4 and then got much improved 
results.  I also enabled pitch & roll optimization, to allow 
compensating for minor errors in camera positioning.  Final ACPD was 
1.103 and rendered image looks very good -- nice straight lines.

Please check this out yourself and see if helps.

--Rik

Terje Mathisen wrote:

> rj.littlefield at computer.org wrote:
>
>> Terje Mathisen wrote:
>> What does "problems trying to find a stable solution" mean?  If the 
>> optimizer converges to different values depending on your initial 
>> estimates, then most likely there are a bunch of local minima in the 
>> error as a function of a/b/c.  I don't think that anything I did 
>> would change the optimizer behavior in that case.
>
>
> I'll probably have to download the source code and try to run it under 
> a debugger myself. :-)
>
>>
>>>
>>> Should I place the photo + my current PTAsm project file on a web 
>>> server for download?
>>
>>
>> Sure, I will be happy to take a look.  But do not get your hopes too 
>> high.  I am not expert on the PT lens model, and it sounds like the 
>> problem may be that the model just cannot represent this lens very 
>> well.  Perhaps someone else with more experience can take a look too.
>
>
> OK, I've placed the source image (8 Mpix, 1.5 MB! :-(), plus two PTAsm 
> project files on my web server:
>
>  http://norloff.org/pano/pano12/
>
> lens-7.1mm.ptp is the regular project, using both horizontal/vertical 
> lines, and several straight line blocks.
>
> The big project file (lens-test.ptp) is the result of running a perl 
> program which takes all the straight line control points and converts 
> them into all possible combinations of first/second point for a 
> horizontal line.
>
> Using the optimizer, I get maximum errors up to 150-200 pixels, which 
> is a _lot_ worse than simply assuming no lens distortion at all. :-(
>
> Terje





More information about the ptX mailing list