Large, complex, scanned imaged mosaic stitch help request.
tcorbet
the subject - without luck, I have tried a variety of products and
can get through parts of a workflow with most of them, but cannot get
all the way to the desired output with any of them. I understand - at
a very low level of understanding - that with PTGui [as with
Photoshop, Hugin and ICE] that most of you are devoted to the more
complex trigonometry required to render DSLR inputs, and that
multiple, flat images from a patchwork of scans is NOT the primary
target of your product and energies.
02. That said, when I downloaded PTGui and simply flailed around
trying to guess what sort of parameters I would have to select to get
something useful, I got close. So, I sent a request to the support
email address, and received some helpful hints, including the pointer
to item 5.16 available in this forum. That's the good news; the bad
news is that following that heuristic yields a completely unusable
result.
03. So, my hope is that if I upload the .pts file from my completely
guessed-at trial [since I have absolutely no way to recreate what the
guessed parameters were, much less what they meant to your algorithms]
and the .pts file that resulted from attempting to follow the 'best
practices' for this type of content, someone will be able to decipher
that binary file to determine what went right and what went wrong in
order to make a suggestion as to what combination of parameters gives
PTGui the highest probability of successfully rendering the composite.
Just for whomever is keeping score, unless I can get PTGui to do it,
it probably can't be done.
A. Hugin has no useful documentation and does not support tiff.
B. ICE has a user interface that is so dependent upon the
machinations of the Raw Camera, Automated Panorama Fixture
Marketplace, that it is not even possible to load the matrix of
patches to be blended.
C. While Photoshop CS5 does improve its memory management, after
recovering the Interactive Mode that they decided to eliminate from
CS3 and CS4, because none of these pieces of software can come close
to automatically determining the correct grid layout [I carefully
align each of the patches using a magnifying glass and usually can let
their 'SnaptoImage' logic pick it up, but sometimes must over-ride
that] -- all of which is quite understandable given the complexity of
the images.], the process abends without any log/diagnostic output.
At least I have been able to get it past running out of virtual memory
problems, but it appears that this task really requires a 64-bit OS
and Interactive Mode support is not available.
Thanks for any support/suggestion you can provide.
Erik Krause
> I have been trying to accomplish a large stitch - as described in
> the subject - without luck, I have tried a variety of products and
> can get through parts of a workflow with most of them, but cannot get
> all the way to the desired output with any of them.
PTGui - as most other panorama software - is dedicated to stitch
panoramas: Shots from the real world taken from one single viewpoint.
Hence the main image parameters all apply to some angles. With two
exceptions: Horizontal and vertical shift.
In theory it should be sufficient to optimize for this two parameters
only, leaving all others to zero (especially Yaw, Roll and Pitch for any
single image on Image Parameters tab and all other lens correction
parameters on Lens Settings tab), but I don't know what maximum values
are allowed for them.
Of course each image needs individual shift parameters, so you need to
check all images but the first (or center one) on Lens Settings tab and
enable optimization for the same images on Optimizer Tab. Disable
optimization for anything else.
You will have to switch to advanced interface to see all this.
Good luck...
--
Erik Krause
http://www.erik-krause.de
John Houghton
> A. Hugin has no useful documentation and does not support tiff.
supported. I just stitched a set of TIFF input images and generated a
TIFF panorama. (I used the version Hugin 2010.2.0 windows x32, but
I'm sure that the latest versions will do likewise).
John
PTGui Support
> 03. So, my hope is that if I upload the .pts file from my completely
> guessed-at trial [since I have absolutely no way to recreate what the
> guessed parameters were, much less what they meant to your algorithms]
> and the .pts file that resulted from attempting to follow the 'best
> practices' for this type of content, someone will be able to decipher
> that binary file to determine what went right and what went wrong in
> order to make a suggestion as to what combination of parameters gives
> PTGui the highest probability of successfully rendering the composite.
Yes that's probably the best idea. Unfortunately this forum doesn't have
an upload facility currently. If you have a website please upload the
pts files, otherwise please send them to sup...@ptgui.com and I'll take
a look.
Joost
tcorbet
file formats; the download that I tried which did not is AutoStitch.
02. I will see if I can find/set the parameters on and off in the
manner suggested and send all three .pts files as Joost suggested.
03. My "non-real-world" challenge -- for anyone else attempting to
work with similar content -- can be viewed here:
http://memory.loc.gov/cgi-bin/addItemLink.pl?tourl=/gmd/gmd416m/g4164m/g4164sm/gpm00001/gpm00001.html&style=gmd&itemLink=r?ammem/gmd:@field%28NUMBER+@band%28g4164sm+gpm00001%29%29
A very brief description of the amazing process by which the
perspective map was drawn appears in "Views and Viewmakers of Urban
America".
While the desire is to capture all 104 patches [which will be a
composite file of about 5.5GB], having failed with some 30 patches
that would keep me under the 2GB limit which I thought was breaking
Photoshop CS3, I have now backed off to 16 patches that have
successfully been [Interactive Mode] stitched by Photoshop CS5 into a
composite that is about 800MB. I can only send a file less than 100MB
on my copy of YouSendIt, but if anyone has any other way to transfer
the larger file, I would be happy to share it with you for viewing.
It's not quite Street View, but it is as close as any of us will ever
come to being able to pan over an area that was probably about a
quarter mile square representing all the buildings and homes present
in St. Louis a decade after the Civil War.
> a look.
>
> Joost
Peter Nyfeler
> I can only send a file less than 100MB
> on my copy of YouSendIt, but if anyone has any other way to transfer
> the larger file, I would be happy to share it with you for viewing.
a try
Best regards
Peter
John Houghton
for the 7-day trial period.
Having looked at the material you are attempting to stitch, it is far
from straightforward. The downloaded images are of different sizes
and suffer from perspective distortion. Nor are they shots of
sections of one large drawing. The images don't contain identical
features in the overlaps. Hence, precision stitching is not to be
expected.
John
On Mar 2, 6:46 pm, tcorbet <terry.cor...@gmail.com> wrote:
> 01. My apologies to those who point out that Hugin does support tiff
> file formats; the download that I tried which did not is AutoStitch.
>
> 02. I will see if I can find/set the parameters on and off in the
> manner suggested and send all three .pts files as Joost suggested.
>
> 03. My "non-real-world" challenge -- for anyone else attempting to
> work with similar content -- can be viewed here:
>
>
> - Show quoted text -
Hugo Rodriguez
I don't know if this has been said here, but I've just saw the result of a
research at the university of Berkeley, where three guys had developed a new
projection for wide angle images that has impressed me:
http://vis.berkeley.edu/papers/capp/
Is there any possibility to implement that in PTGui?
Hugo Rodriguez
Erik Krause
> I don't know if this has been said here, but I've just saw the result of a
> research at the university of Berkeley, where three guys had developed a new
> projection for wide angle images that has impressed me:
>
> http://vis.berkeley.edu/papers/capp/
Well, that's from 2009.
> Is there any possibility to implement that in PTGui?
I think it should be possible to tweak Vedutismo to give equal or better
results. See paper with comparisons on
http://vedutismo.net/Pannini/
Although I must admit that the idea of marking lines which should stay
straight is compelling....
John Houghton
John
On Mar 3, 7:32 pm, "Hugo Rodriguez" <info.hugorodrig...@gmail.com>
wrote:
Roger D. Williams
That was a good tip, Peter. Thanks. I can use this. It nicely links
my iPhone, Notebook and Desktop computers. I see they insist on
being able to send me advertising, not only their own, but I have
used a mail account I keep for this kind of communication.
I don't consider this a thread hijack because it is still concerned
with handling the transfer of large files, although I must admit it's
strayed a little from that theme.
Roger W.
--
Work: www.adex-japan.com
Hugo Rodriguez
Hugo Rodriguez
Enviado desde mi telefono
> --
> You received this message because you are subscribed to the Google Groups "PTGui" group.
> To post to this group, send email to pt...@googlegroups.com
> To unsubscribe from this group, send email to ptgui+un...@googlegroups.com
> Please do not add attachments to your posts; instead you may upload files at
> http://groups.google.com/group/ptgui/files
> For more options, visit this group at http://groups.google.com/group/ptgui
Erik Krause
> It's really annoying to have threads hijacked in this way.
The problem is that most people don't even know they're doing it. They
never heard of threaded reading, In-Reply-To or References headers and
all that. And especially Microsoft doesn't make it easy to learn this
either.
Hugo, apparently you pressed "Reply" while reading John's mail about
"Large, complex, scanned imaged mosaic..." and changed subject instead
of pressing "New Message". "Reply" places a header into the message with
the unique ID of the message you are currently reading. This enables
(better) mail readers to put the message in thread context, f.e. display
them in a tree view, which is a great help to read high traffic lists
like PTGui or PanotoolsNG.
Hugo Rodriguez
Thanks for the help. Now I understand what's happened. It's exactly what you
explained:
1) In my Outlook, I replied to another message. I didn't look for any,
simply replied to the first one I have in front of me.
2) Deleted the WHOLE message and the subject, and wrote the new one.
So, I apologize for the inconvenience. I've learned from that. Please accept
my excuses.
I have to say that I've done this sometimes on other forums and never had a
problem. In fact, I've just asked some friends and they told me they often
do exactly the same.
Anyway, John, I believe that even that is anoying, it's not so bad as to say
I'm hijacking the thread. I've just wanted to share what I think is a very
interesting info with all the community. If I enter the google forum page
and see something like this, I quickly understand (as I'm seeing now) that
it's a mistake of some member related to management of the messages, not a
bad intention.
Hugo Rodriguez
-----Mensaje original-----
De: pt...@googlegroups.com [mailto:pt...@googlegroups.com] En nombre de Erik
Krause
Enviado el: viernes, 04 de marzo de 2011 12:18
Para: pt...@googlegroups.com
Asunto: [PTGui] Re: Large, complex, scanned imaged mosaic stitch help
request. Options
--
John Houghton
wrote:
> I'm hijacking the thread.
mosaic stitch help request." vanishes from the list of discussions
that I see in my Google Groups window, and your new one appears
instead. Anyone following the original discussion then has to hunt
for the thread.
John
Erik Krause
> I have to say that I've done this sometimes on other forums and never had a
> problem. In fact, I've just asked some friends and they told me they often
> do exactly the same.
Yes, it's a very common. And it might play no role in a low traffic
list. But if you follow 5 lists with 30 to 50 messages per day each you
need some structure. BTW.: It's a drawback for the poster actually,
since if he replied in a thread which many users choose to ignore he
might get no attention...
> Anyway, John, I believe that even that is anoying, it's not so bad as to say
> I'm hijacking the thread.
Thread hijacking is the common term for this:
http://www.google.com/search?q=Thread+hijacking
Joergen Geerds
posts from the list (sanitize this thread) via the "More options" link
for each post in the google groups interface.
Then Hugo can create a new thread talking about the Carroll
projections, or find my post from ages ago, and reply to that one (I
think I had posted about the Carroll projections before as a feature
request).
joergen
Hugo Rodriguez
Again, I'm sorry for my mistake. Indeed it's difficult to follow threads
with so much messages.
So I don't want to continue placing my wrong thread into another. So this
will be my last message in this thread.
Regards,
Hugo Rodriguez
-----Mensaje original-----
De: pt...@googlegroups.com [mailto:pt...@googlegroups.com] En nombre de Erik
Krause
Enviado el: viernes, 04 de marzo de 2011 16:25
Para: pt...@googlegroups.com
Asunto: [PTGui] Re: Large, complex, scanned imaged mosaic stitch help
request. Options
--
ozbigben
> 03. My "non-real-world" challenge -- for anyone else attempting to
> work with similar content -- can be viewed here:
>
> A very brief description of the amazing process by which the
> perspective map was drawn appears in "Views and Viewmakers of Urban
> America".
>
try downloading some of them and see what I can do.. it may be useful
if we have something similar in our archives.
Firstly, looking at the software you mentioned I'd say PTGui is
probably the most likely to give you a successful result. Photoshop
and ICE are OK when they work, but they're a pain (if not impossible)
when you need to tweak individual images. I'm working with scans of
old maps around 800Mb each and no program I've tried can auto-align
them (no errors, but no alignment either). While theoretically you
only need to adjust the offset, in practise, some viewpoint correction
can sometimes be handy.
Normally I'm only stitching 2-4 images (e.g.
http://files.digitisation.unimelb.edu.au/zoomify/melbourne_1880.html),
but I use a slightly different approach to some others. Rather than
justoptimise offset, I set the focal length of the "lens" to something
ridiculously large (1000mm for 2-4 images) and optimise yaw, pitch and
roll. Let's just hope the images are all the same width, and apart
from that, as these ar individual drawings they may not necessarily
line up that well
Ben
ozbigben
The errors are quite substantial in places, and they don't line up as
a grid as you probably hoped. Adding more images may change the
alignment to more of a grid but this would also increase the errors in
individual seams. The attention to detail in these drawings is
impressive, but there is the occasional discrepancy between images
where buildings in the overlap are completely different. (middle right
in this example)
I've uploaded the pts and resulting stitched image at http://bigben.id.au/demo/st_louis/
I renamed the images to their tile number, but also left the original
name in there as well. As the images are not cropped to a fixed size
you need to vary the fov for each image (except the first). Optimising
viewpoint in this case makes no substantial difference. Control point
distances in this case is (average: 20'ish, Min: 6, Max: 66
Ben
ozbigben
Having said that it won't work as a "seamless" stitched image doesn't
mean that it wouldn't still make an interesting composite image. You
could align all the images as best as possible and then create a multi-
layered PSB of the layers. Then manually crop each layer to the image
and give it a thin stroke around the edge... and maybe even a slight
drop shadow.
ozbigben
and will post the PTS file when it's done. I've completed one row and
it looks like it will work reasonably well. Going through this process
also shows you a few things about the original images. The attention
to detail in terms of matching the overlap between images is quite
remarkable in some areas with even people in the streets being drawn
in the same place in adjoining images. Towards the outer images
though, things tend to get a bit more lax, with significant variation
in streets and buildings, making control point selection difficult.
There is also a greater variation towards the top and bottom of an
image compared to the centre. In some cases (as with my first test)
this can throw things out a bit. In the control point window, link
the image scrolling and keep auto add turned on. When working your
way across a row, start adding control points near the centre of the
image and then work your way up and down. After you add 2 control
points, PTGui should start trying to auto locate matching locations to
subsequent control points. If these miss the mark by a lot, have a
close look at the two images and consider ignoring that match.
Some of the images also have very little overlap making it a little
harder to find matching features. This would also have prevented auto
alignment from wokring even if the drawings were accurate enough.
There are also a few odd corners that are drawn as insets. These can
also be placed in the final composite by listing the image twice in
the panorama and doing a custom crop to isolate the inset.
tcorbet
What was your starting point for the Melbourne example? Did you have
access to a single, large photo/line-drawing covering the entire
scene, or as in the St. Louis case, did you have discrete, overlapping
patches? If you had access to the whole plate/canvas, what was its
size? If you started from a whole plate/canvas, did you scan/digitize
in 'chunks'?
Where I seem to see a ragged, almost 'torn' edge line between the left
and right portions of the composite, was that something you created in
your approach, or is that in the source images?
On Mar 4, 6:46 pm, ozbigben <ozbig...@gmail.com> wrote:
> OK, getting back to the topic ;-)
>
> > 03. My "non-real-world" challenge -- for anyone else attempting to
> > work with similar content -- can be viewed here:
>
> >http://memory.loc.gov/cgi-bin/addItemLink.pl?tourl=/gmd/gmd416m/g4164...
>
> > A very brief description of the amazing process by which the
> > perspective map was drawn appears in "Views and Viewmakers of Urban
> > America".
>
> This is the kind of challenge that pops up in my current job. I'll
> try downloading some of them and see what I can do.. it may be useful
> if we have something similar in our archives.
>
> Firstly, looking at the software you mentioned I'd say PTGui is
> probably the most likely to give you a successful result. Photoshop
> and ICE are OK when they work, but they're a pain (if not impossible)
> when you need to tweak individual images. I'm working with scans of
> old maps around 800Mb each and no program I've tried can auto-align
> them (no errors, but no alignment either). While theoretically you
> only need to adjust the offset, in practise, some viewpoint correction
> can sometimes be handy.
>
ozbigben
A0 in size, but our scanner only handles A1 so it was scanned in 2
chunks and then stitched. This particular image is a reproduction of
the original, so the ragged lines are in the reproduction. There is a
small area of mis-registration in the centre if you look really
carefully, but it was just a quick demo so I haven't got around to
fixing it. If you zoom in all the way using the slider, you can see
the start of the misalignment at the top centre. It extends upwards
from for a short distance and is largely due to the original not lying
completely flat (even though it was under glass).
Steve Kim
Thank You.
Best,
Steve Kim.
2011년 3월 4일 금요일 오후 9시 34분 58초 UTC-6, ozbigben 님의 말: