gui idea: masking for better CP generation

[Zoran Zorkic]

I don't know if it's been mentioned but here's a feature I'd love to
see:

* insert photos
* run a CP generator & optimize for preliminary alignment
* open up a V window with an appropriate projection
* mask areas bad for CP generation
* rerun CP generation using masks (masks being automagicaly applied to
the individual photos) to avoid bad CP

Rejoice at not having to pick CP's manually. Or so I hope :)

 I think Hugin would benefit greatly from some robust masking tools.

What do you think?



--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[sebastien delcoigne]

Following the masking feature mentioned by Zoran, a masking feature to tell which part of a photo should be discarded (in the case of moving objects for example), with the possibility to mask directely in the hugin interface. Would something like this be possible ?
It would enabled to keep the entire workflow within hugin (today's method : http://hugin.sourceforge.net/tutorials/enblend-svg/en.shtml)

On Thu, Oct 22, 2009 at 7:30 AM, Zoran Zorkic <zomba@...> wrote:

I don't know if it's been mentioned but here's a feature I'd love to
see:

* insert photos
* run a CP generator & optimize for preliminary alignment
* open up a V window with an appropriate projection
* mask areas bad for CP generation
* rerun CP generation using masks (masks being automagicaly applied to
the individual photos) to avoid bad CP

Rejoice at not having to pick CP's manually. Or so I hope :)

 I think Hugin would benefit greatly from some robust masking tools.

What do you think?

--
Sébastien

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Zoran Zorkic]

This workflow I mentioned, uses masking to control CP generation, but
the same masking tools would help with the classic use of masking to
control final blending.

Masking tools should have a fast coarse mode, so users can get results
fast, and of course precise tools for fine tuning.

I got this idea after spending too much time adding CPs on photos with
lots of sharp edges and contrasty areas which should be perfect for
CPs. Alas both APSc and PanoMatic manage to miss them and put CPs on a
blank wall and such. Get's tedious real quick.

Masking on all photos at once should really speed things up, and it
should not be than hard to implement, at least for CP generation.
For blending use, masking would have to be more complicated and more
thought out.

On Oct 22, 9:50 am, sebastien delcoigne
<sebastien.delcoi...@...> wrote:

> Following the masking feature mentioned by Zoran, a masking feature to tell
> which part of a photo should be discarded (in the case of moving objects for
> example), with the possibility to mask directely in the hugin interface.
> Would something like this be possible ?
> It would enabled to keep the entire workflow within hugin (today's method :http://hugin.sourceforge.net/tutorials/enblend-svg/en.shtml)
>
>
>
> On Thu, Oct 22, 2009 at 7:30 AM, Zoran Zorkic <zo...@...> wrote:
>
> > I don't know if it's been mentioned but here's a feature I'd love to
> > see:
>
> > * insert photos
> > * run a CP generator & optimize for preliminary alignment
> > * open up a V window with an appropriate projection
> > * mask areas bad for CP generation
> > * rerun CP generation using masks (masks being automagicaly applied to
> > the individual photos) to avoid bad CP
>
> > Rejoice at not having to pick CP's manually. Or so I hope :)
>
> >  I think Hugin would benefit greatly from some robust masking tools.
>
> > What do you think?
>
> --
> Sébastien

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Bruno Postle]

On Wed 21-Oct-2009 at 22:30 -0700, Zoran Zorkic wrote:
>
>I don't know if it's been mentioned but here's a feature I'd love to
>see:
>
>* mask areas bad for CP generation
>* rerun CP generation using masks (masks being automagicaly applied to
>the individual photos) to avoid bad CP

The preferred interface for specifying input to control point
generators is to supply a .pto project.  There would be no problem
also supplying any masks in this project file, the control point
generator would have to be able to read them though.

--
Bruno

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Oskar Sander]

2009/10/22 Bruno Postle <bruno@...>

The preferred interface for specifying input to control point
generators is to supply a .pto project.  There would be no problem
also supplying any masks in this project file, the control point
generator would have to be able to read them though.

Have there ever been a discussion on showing dependencies in the pto files to help CP-generators.  I.e. "image 1 overlaps image 2,3,4" without specifying positions?

/O


--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Yuval Levy-6]

Zoran Zorkic wrote:
> * run a CP generator & optimize for preliminary alignment

why?


> * mask areas bad for CP generation

in principle this is a good idea and has been discussed a number of
times. the last time I had a thorough discussion about it was with Tim,
when mentoring him about Celeste.

One original idea for Celeste was to generate masks for the sky as one
such "bad areas". The masks would be useful also for other things such
as post-processing (common use case: apply different curves to the sky).


> * rerun CP generation using masks (masks being automagicaly applied to
> the individual photos) to avoid bad CP

it makes sense to *run* CP generation right away the first time using
masks that maps out good areas / bad areas.

Normally the best areas are those that don't move and are far away from
the camera, and this does not depend on on pre-alignment - hence my
above question, why run a CP generator & optimize for preliminary alignment?

IMO preliminary alignment should be much faster than running a fully
fledged CP generation. Use an heuristic approach with the data available
intrinsically.

First: if the user provides a template, use it.

Then, based on EXIV data determine whether there are multiple exposures.
- pattern? this is HDR shooting
- random? auto-exposure
- constant? plain vanilla panorama

Then, based on EXIV data and quantity of images guess the type of coverage
- small partial panorama
- possibly a cylinder
- maybe the whole sphere
- or there can be way more material than necessary to cover 360x180

Go on making such educated guesses (they are cheap). Then, based on the
guesses, run a few random CP detections between a couple of likely to be
adjacent pairs. This will tell you, their relative positions, and
improve the guesses from wild to likely until the probability is very
high that the guess is right.

Once the guess is right (e.g.: there are 90 images in 3 exposure layers
and they are arranged 6x5) spread their initial positions.

Optional: pass your guess by the user (visualize it and ask him if it
makes sense).

Go on generating the CPs, pairwise, per adjacent pictures.

Of course the heuristic approach must make provisions for things that
can go wrong:
- maybe there is no usable order (e.g. hand held, shot wildly)
- maybe there is an outlayer (e.g. starting to shoot a sphere, then
moved rapidly to shoot a plane as it was flying by, then continued the
sphere)
- maybe once the bad areas are masked (e.g. crowds, clouds, leaves,
close-by objects) there are no more features in the picture to generate CPs

in all such cases, raise an exception, ask the user's for manual
intervention.


Yuv

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Bruno Postle]

On Fri 23-Oct-2009 at 12:37 +0200, Oskar Sander wrote:
>
>Have there ever been a discussion on showing dependencies in the pto files
>to help CP-generators.  I.e. "image 1 overlaps image 2,3,4" without
>specifying positions?

Specifying approximate positions should be fine, the control point
generator could just look at the angular distances and angles
of view to get a good idea which photos are likely to overlap.

The gsoc2009_layout branch does support a 'j' parameter to indicate
likely stacks, but this isn't used by anything yet either.

--
Bruno

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Rogier Wolff]

On Fri, Oct 23, 2009 at 07:38:46AM -0400, Yuval Levy wrote:
> > * rerun CP generation using masks (masks being automagicaly applied to
> > the individual photos) to avoid bad CP
>
> it makes sense to *run* CP generation right away the first time using
> masks that maps out good areas / bad areas.
>
> Normally the best areas are those that don't move and are far away from
> the camera, and this does not depend on on pre-alignment - hence my
> above question, why run a CP generator & optimize for preliminary alignment?

Because it allows you to have a preliminary idea about where the
overlaps are.

I think I ran "align image stack" a while ago, and it tries to generate
CPs in equal amounts in 25 sub-areas of the image. This gives a much better
control point spread. Each sub-area is therefore about 400k pixels.

Having a preliminary idea about the layout, you can now say: images 1 and 2
overlap on an area of about 2 megapixels. You can subdivide that into 5 areas
of 400k pixels, and try to find say 5 controlpoints in each. Matching those
will find only control points in the corresponding 400k pixel area in image 2!

Thus the matching phase will be quite quick!  It is acceptable if there is
little contrast in a whole area, to have no controlpoints at all. Not to
be preferred, but acceptable.

> Go on making such educated guesses (they are cheap). Then, based on the
> guesses, run a few random CP detections between a couple of likely to be
> adjacent pairs. This will tell you, their relative positions, and
> improve the guesses from wild to likely until the probability is very
> high that the guess is right.

This too would reduce the time needed for CP matching!

My suggestion is to simply match each image with its neighbors.
This is "cheap", and may provide a preliminary layout.
Suppose I shoot a snake patterned multi-row pano as below. Then
image 6 will be next to image 5, and then image 7 will be below
image six. This provides a hint that 8 and 5 might overlap as
well....

   1  2  3  4  5  6
    ... 10  9  8  7


In the case that you shoot bracketing HDR, you'll find that
the stacks line up with very little difference between them, but
still you'll be able to make a "snake" like before!

> Once the guess is right (e.g.: there are 90 images in 3 exposure layers
> and they are arranged 6x5) spread their initial positions.
>
> Optional: pass your guess by the user (visualize it and ask him if it
> makes sense).
>
> Go on generating the CPs, pairwise, per adjacent pictures.
                                          ^^^^^^^ overlapping.


        Roger.

--
** R.E.Wolff@... ** http://www.BitWizard.nl/ ** [phone] **
**    Delftechpark 26 2628 XH  Delft, The Netherlands. KVK: 27239233    **
*-- BitWizard writes Linux device drivers for any device you may have! --*
Q: It doesn't work. A: Look buddy, doesn't work is an ambiguous statement.
Does it sit on the couch all day? Is it unemployed? Please be specific!
Define 'it' and what it isn't doing. --------- Adapted from lxrbot FAQ

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Oskar Sander]

2009/10/23 Bruno Postle <bruno@...>

On Fri 23-Oct-2009 at 12:37 +0200, Oskar Sander wrote:
>
>Have there ever been a discussion on showing dependencies in the pto files
>to help CP-generators.  I.e. "image 1 overlaps image 2,3,4" without
>specifying positions?

Specifying approximate positions should be fine, the control point
generator could just look at the angular distances and angles
of view to get a good idea which photos are likely to overlap.

Or XYZ + angular distances + angle of view in the case of mosaic...
 

/O

--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---

[Zoran Zorkic]

On Oct 23, 1:38 pm, Yuval Levy <goo...@...> wrote:

> Zoran Zorkic wrote:
> > * run a CP generator & optimize for preliminary alignment
>
> why?
>
> > * mask areas bad for CP generation
>
> in principle this is a good idea and has been discussed a number of
> times. the last time I had a thorough discussion about it was with Tim,
> when mentoring him about Celeste.
>
> > * rerun CP generation using masks (masks being automagicaly applied to
> > the individual photos) to avoid bad CP
>
> it makes sense to *run* CP generation right away the first time using
> masks that maps out good areas / bad areas.
>
> Normally the best areas are those that don't move and are far away from
> the camera, and this does not depend on on pre-alignment - hence my
> above question, why run a CP generator & optimize for preliminary alignment?

The idea was to have as little manual work as possible. For it to be
possible you need aligned photos, but the point is to draw the mask
for all photos at once and have Hugin unwrap and apply the mask to
individual photos.
For that to work you need to have you photos aligned.
For my use case, d90+18mm+nodal ninja, that workflow would help a lot,
as I get a good alignment from the start, but have to spend a lot of
time to get perfect alignment :(



--~--~---------~--~----~------------~-------~--~----~
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 hugin-ptx@...
To unsubscribe from this group, send email to hugin-ptx+unsubscribe@...
For more options, visit this group at http://groups.google.com/group/hugin-ptx
-~----------~----~----~----~------~----~------~--~---