On 9/8/14 7:01 PM, Bruno Postle wrote:
Hello again,
I've been doing some tests -- many with an alternate tool, so
I'm back to learn more.
First, the alternate tool I've been testing is the Gigapan Stitcher
tool, and since I'm using a Gigapan panorama head it seemed like a
natural -- except for my predilection for Linux and open-source that
is.
So the stitch that took Hugin 6 hours (and for which I hadn't found
the proper means of setting the scope of the images), took Gigapan
Stitcher 16 minutes. And that was with basically one button press:
-> "Stitch"
The downside is that it too of course makes some errors, and I don't
have the option to go in and tweak things. Well, I can tweak the colors,
but I can't mask parts from particular images, and force the inclusion
from others. I can't go in and tweak control points when it makes
mistakes.
So the two panoramas I've stitched with Gigapan stitch can be viewed
at:
- http://web.avl.indiana.edu/~shermanw/MST
the second one is from 1547 images, but if you look to the left,
you'll notice that that process went a little wonky. It's actually
a cool effect -- love that AT&T building. But for the Gateway Arch,
it really could have done a much better job.
The other downside of Gigapan stitch is that it can't handle hand-held
shoots unless they rigidly take the same number of pictures in each
column & row.
>> My first issue is that I'd like to find a way to constrain what images
>> are compared in the control-point search. I've taken a range of
>> panorams, and the largest are quite large -- over 1000 pictures. So
>> I'd like to find a way to limit control point analysis to say the 8
>> neighboring pictures of any one picture.
>
> You can try different control point strategies in File -> Preferences.
> The default CPFind detector tries to minimise the number of comparisons
> by assuming you have taken photos in rows. Alternatively if you can
> approximately arrange the photos using a template, there is a
> 'prealigned' detector that only compares photos that are nearby.
Nice, glad to learn about that. One things makes my wonder -- after
choosing "Hugin CPFind (prealigned)" and double-clicking on it, there
is a setting window, and for Type, it has "All images at once", and
not "Prealigned panorama". Is this a bug, or some orthogonal setting?
>> Okay, so now onto a 210 image shoot (42x5) using a Gigapan robot.
>>
>> First, loading the images it put them horizontally, whereas they were
>> shot vertically, so it has a typical striation pattern from a
>> misalignment -- I know that at this stage that doesn't matter, but
>> would be nice if I could provide Hugin with the basic layout of the
>> images.
>
> Hugin will read any EXIF orientation data from the camera to try and get
> this right. If your photos don't have this data then you need to used
> the advanced/expert interface to set the 'roll' of the photos before
> alignment.
Right, well that's not exactly what I meant. What I meant is that if
I shoot pictures in columns -- say 5 pictures per column, and then
import them into Hugin, it lays them out going left to right, not top
to bottom, so they are in the wrong place. Of course I wouldn't know
how to tell Hugin how many pictures per column, so I'd have to know
that too. Anyway, that matters only once I figure out how to do
the prealigned option.
>
>> I then ran the "2. Align" step. And this took over 6 hours to process.
>> It was not fun waiting for that, and I dread what will happen when I
>> get to my "large" image collections! I did notice that at times all
>> the CPUs were going, and then other times just a single CPU. Also, I'd
>> be interested to know exactly what "Optimizing Variables" were being
>> optimized to get a sense for where it is in the process.
>
> Definitely you need to stop using the Assistant for such a big image
> set. The Assistant will optimise all sorts of lens parameters by
> default, whereas if you have already characterised your lens there is
> only any need to optimise roll, pitch and yaw for the photos.
>
> The optimiser is much quicker when optimising fewer parameters.
Okay, good. And this is where my past experiences with Hugin may
hinder me (or maybe not), but when I go to the "Advanced" layout, under
Photos it has the name, and then columns for y, p & r, whereas from
the last I remember lots and lots of parameters for each image, which
could be turned on and off for optimization.
Aha, found a few more by switching to "Expert" mode. (I previously
hadn't noticed a difference between "Advanced" and "Expert".)
Hmmm, but I don't see the buttons that I thought were there to
tell the optimizer to skip certain values.
>> Eventually, it produced a result, and the pictures are in the proper
>> order.
>>
>> With one huge problem -- the pictures wrap around past 360 to about
>> 400 degrees. In reality the shoot was about 270 degrees.
>
> The angle of view parameter for your _photos_ is too high. Normally the
> Assistant optimises this, but if your panorama doesn't cover much of the
> sphere surface then there is no way of knowing how big it is supposed to
> be - i.e. your panorama is too thin and incomplete to be able to
> calculate lens angle of view.
>
> You need to characterise the angle of view of your lens by aligning a
> full 360°, then use this number from then on. You could alternatively
> stitch a not-thin panorama, say 90x90°, this would work for calculating
> lens angle of view too.
I guess this seems like a problem to me. First, I compiled the "lensfun"
tool, does that not have a database of lenses to at least get it
in the ballpark of what the parameters are -- which can then be
fine tuned by the optimizer?
I'll go back to my Gigapan Stitch experience -- I clicked "Stitch",
and if basically figured out that the sum total was about 270deg
horizontally. I suppose it may have used a lens database to figure
that out.
In hugin, under the "Photos" tab, there is a "Lens type" section,
and it has values for Focal length and Focal length multiplier.
Does this map somewhat at least to a DSLR lens type? So if I have
a 50mm prime on a full-sensor camera, can I provide a good guess
as to the parameters -- actually, for my 210 picture shoot it's
200mm on a 70-200 lens.
Ahh, just found a bunch more parameters for each photo with the
radio buttons to the right of the image list letting me view
lens parameters and photometric parameters, etc.
But I don't see how to set an anchor image.
So I guess, am I supposed to select each image and set it to
a new (smaller) value of "Hfov(v)".
Okay, so pardon the stream-of-consciousness flow of this post,
but maybe this will be helpful to anyone looking to understand
the newby user's experience.
So I did alter the "Hfov(v)" for one image, and that affected
all the images -- so obviously all the images point to the same
lens parameter entry.
BUT, lowering the value from 17.1 to 10.0 didn't affect the
fact that the panorama still things it's a 400 degree shot.
I then tried 5.0, and no change. So I don't know where I'd
put the lens parameters that would inform Hugin of how the
images should be laid out.
I looked and looked for a way to correct this -- tried using the
"Field of View" parameter under the "Projection" tab, and pressed the
"Fit" button, but that just did some processing, then returned the
horizontal FOV to 360, but shrunk the vertical FOV! But it left all
the pictures where they were, so there is still about 40 degrees of
overlap.
This is adjusting the angle of view of the _output panorama_, which can
never be larger than 360°.
Right!
I figured I might as well see what happens
Segmentation fault (core dumped)
make: *** Deleting file `IMG_7374-IMG_7584_fused.tif'
I looked at the "stack_ldr" pictures, and as I expected the two ends
of the panoram are blended together.
Yes this is going to be a mess because you have overlapping photos that
shouldn't be.
It's not the image I put up on my page, but when I did get it
to provide an output, it was an interesting picture, and most
of it was fine.
I also noticed that the list of "exposure_layers" temporary images was
missing at least one in the sequence, so perhaps this could be related
to the seq-fault.
The segfault could be something else, possibly just running out of memory.
One thing I notice is that for my small panoram, the Batch Processor
is informing me that a Cylindrical projection was chosen for the
output, and for the medium, a Mercator projection was chosen.
You need to take control of the software. The Assistant will pick a
projection for you using heuristics, but you probably always want to
stitch these panoramas in equirectangular projection.
In general, Hugin has lots of options and widgets which can be
bewildering, the feature-matching, alignment and stitching process also
has some quirks which are logical but unexpected until you see them in
action - trying to learn all this with a large image set is going to be
extremely frustrating.
And so I've looked at some of the tutorials, but of course with the
major change in the GUI, many of the tutorials are for the old version.
I guess one thing that would be handy is for the names of each tutorial
to include what version of Hugin they were created with. I haven't
yet found that one tutorial that I need.
I did notice that Terry has a new tutorial on stitching mosaics which
uses the new interface (with Hugin 2013). But most of the rest of
them are pre-2013.
Is there a method by which I can process panorams consisting of 1600
images with Hugin? I sure hope so.
Yes though you need lots of memory for stitching.
I have 24 Gig of RAM on this machine -- and the machine that sticked
the 1547 image panorama had only 12 Gig. And that took about 6 hours.
Oddly, it took like 3 hours to write out the 12Giga-pixel image in PNG
format. Only took like 30 minutes to write as a Big-TIFF.
Anwyay, I'll press on. I just wanted to provide some feed back from
a newby perspective. Not trying to be unnecessarily critical, because
I love that this program is here, but I want to love it more.
Thanks again,
Bill
--
A list of frequently asked questions is available at:
http://wiki.panotools.org/Hugin_FAQ
---
You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion on the web visit
https://groups.google.com/d/msgid/hugin-ptx/54114DF4.4040904%40indiana.edu.
For more options, visit https://groups.google.com/d/optout.
