Thanks Yuv for your very detailed response!

The project did use a pano head, so all stacks will be aligned with
the exception of the nadir, which may have some slight movement, so I
add control points to align the Nadir and specify a different lens
after the intial alignment.  I am still using autopano-sift-c.  Yes, I
use a lot of control points then delete many during the processes of
successive optimizations.  I can usually acheive about 0.3 pixel mean
error for a 16k wide pano.

For this project, the intital alignment looked fine.  It was when I
removed the outlier control points and realigned when I ran into
problems.  Seems like my choice for optimization parameters may have
caused the problems.  Let me try it again when I finish my current
project.  Should be able to report back in a few days.  If I continue
to have problems, I will open a bug report.  Thanks again for your
very detailed response!

Regards,
Rick

On Dec 27, 12:39 pm, Yuval Levy <[email protected]> wrote:
> Hi Rick,
>
> On December 26, 2010 09:33:28 pm RueiKe wrote:
>
> > My attempt to use the latest build on a typical complex project
> > indicates significant issues.
>
> I downloaded your PTO file but did not have time to look at it right away and
> now I don't have the "significant issues" you mentioned present.  However I
> played with it and found no "significant issue".
>
> The project loaded well (with placeholder images, that is).  The application
> is responsive.
>
> I see 31 stacks defined.  I see a lot of CPs.  I don't recall how you
> generated them; nor can I visually judge their quality.  I find the quantity
> excessive - personally I would do such a project with less than 200 CPs, while
> you have more than 9000.  Even if this was about stitching 217 individual
> frames (single exposures; or misaligned stacks), 1000-1500 CPs would be more
> than enough.
>
> I assume that the stacks are perfectly aligned (shot on a tripod or so), and
> that you want to tonemap a resulting HDR.  Have you considered merging each of
> the stacks first into an HDR image first using pfstools or LuminanceHDR?  This
> would cut computational work by as much as 80%.
>
> It would be helpful if you could detail the issues in ticket(s) on the bug
> tracker so that when trying to figure them out, developers don't have to
> scramble around for information.
>
> I am working on this in my spare time.  I keep folders on my HDD - one folder
> per issue - and the name of the folder is usually the number of the tracker
> ticket (but in this case it was your pseudonym because there was no tracker
> ticket).  It takes only a few seconds to be back on the right page that way.  
> It takes much longer if there is no ticket.
>
> > I understand that my use case of 217 images is not typical
>
> It may not be typical in the sense that it is not the average project;  and
> that probably there is a more efficient way to deal with the project; but it
> is not unique in terms of size.  There are other users processing that amount
> of images and more.
>
> > , but I think it has always been a critical feature for hugin to be able to
> > handle very large projects.
>
> "critical feature" is a bit of an inflated qualification.  There has never
> been any kind of warranty, implied or explicit, that Hugin is fit for any
> purpose.  It is true that Hugin has been handling successfully projects of
> this size and larger in the past; that we take seriously reports such as yours
> and try to help you get the most out of Hugin.  But a single report from a
> single user is never critical until confirmed.
>
> If the report was in the bug tracker, other users experiencing the same issue
> could confirm; add their own experiences and observations; and we would get a
> better picture of whether this is critical or not and what we are dealing
> with.
>
> The basic assumption for now is that this is a specific issue of your specific
> situation.  And with all due respect, even if you say that this happens to you
> in a number of projects (like Milkman with the Samyiang lens),  the basic
> assumption stays that it is an issue of your setup and not of the Hugin code. 
>  
> To be considered an issue with the Hugin code, it must be confirmed by more
> than one user.  This is what we have triagers working on the bug tracker for.
>
> > I have gone back to 2009.4 with no issues.
>
> That's a long way back.  I would be surprised that an issue is hiding in the
> code for one year with no report.  You will forgive me for thinking that the
> issue is on your end rather in the Hugin code until proven different.
>
> > Let me know if there is any other
> > testing needed for the issue I described the the RC1 thread.
>
> I see that the optimizer is set to optimize custom parameters.  The choice of
> custom parameters is to optimize one exposure for each stack; and for each of
> those exposures it is set to optimize y/p/r/X/Y/Z.  Plus the lens parameters. 
>  
> all in one go.  This is calling for troubles.
>
> I assume this is a panorama, not a mosaic.  2009.4 works for you because it
> did not have X/Y/Z so it did not let you optimize for those parameters that
> need no optimization.
>
> Indeed, I get what seems to be very reasoable results.
>
> To do so, I first optimize only y,p,r,v in a first run.
>
> Optimizing the lens distortion parameters before the images are properly
> distributed is risky since the optimizer can run into a bad local minimum.
>
> In general I would recommend that you calibrate your lens once and then use
> those calibrated values for the v/a/b/c values.  Or not optimize for a & c at
> all.
>
> I also see that you have two lenses defined.  I assume it is for the nadir?  I
> would optimize nadir position and lens in separate optimization runs.
>
> My ageing Athlon 64 X2 6000+ with 8GB RAM processed the optimizations in a few
> minutes while staying very responsive, playing an internet radio in the
> background and letting me play backgammon and write this message.
>
> I can not verify the results visually, and did not go into much more detailed
> optimization (e.g. using X/Y/Z for the nadir only, but only if the CPs
> connecting the nadir to the other images are on a flat surface; and by
> temporarily transforming the pano so that the nadir is at the center of the
> equirect, on the horizon.
>
> I am sure better results can be obtain with pruning and further fine tuning.  
> But overall I see no problem.
>
> Attached is my (blind) result.  Please confirm if it is garbage or not, and if
> you can improve on it by pruning CPs, etc.
>
> I hope this feedback helps you make sense of 2010.4 and that we can see more
> of your creations on Flickr.
>
> Yuv
>
>  dummified_DR13_Rev1.pto
> 638KViewDownload
>
>  signature.asc
> < 1KViewDownload

-- 
You received this message because you are subscribed to the Google Groups 
"Hugin and other free panoramic software" group.
A list of frequently asked questions is available at: 
http://wiki.panotools.org/Hugin_FAQ
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to 
[email protected]
For more options, visit this group at http://groups.google.com/group/hugin-ptx

Reply via email to