I have commited my latest changes for MatchPoint, suite for detection
of control points(GSoC remember?). The suite is now somewhat stable
and can be used as a replacement for generatekeys from existing
autopano application. I would like someone to give it a try,
preferably someone who has experience with existing autopano
application. In the svn folder there are also 2 images for testing
purposes.
There is no gui, so those willing to test it, will have to:
- compile it using cmake& make(I can provide linux binaries on
request)
- invoke the app from the command line
- [optional] use -t argument to output file used for descriptor
performance testing with matlab(Mikolayczik suite)
Usage:
# first extract features from the first image and output them to the
image1.key:
MatchPoint image1.jpg image1.key
# for second image:
MatchPoint image2.jpg image2.key
#match features from the two generated files using autopano:
autopano project.pto image1.key image2.key
# open the project file in hugin:
hugin project.pto
Notes:
- points close to image edges are not detected(need to improve this)
- creating descriptors is a bit slow(see TODO item in Descriptor.cpp
how to fix this)
> I have commited my latest changes for MatchPoint, suite for detection > of control points(GSoC remember?). The suite is now somewhat stable > and can be used as a replacement for generatekeys from existing > autopano application. I would like someone to give it a try, > preferably someone who has experience with existing autopano > application. In the svn folder there are also 2 images for testing > purposes.
> There is no gui, so those willing to test it, will have to: > - compile it using cmake& make(I can provide linux binaries on > request) > - invoke the app from the command line > - [optional] use -t argument to output file used for descriptor > performance testing with matlab(Mikolayczik suite)
> Usage: > # first extract features from the first image and output them to the > image1.key: > MatchPoint image1.jpg image1.key
> # for second image: > MatchPoint image2.jpg image2.key
> #match features from the two generated files using autopano: > autopano project.pto image1.key image2.key
> # open the project file in hugin: > hugin project.pto
> Notes: > - points close to image edges are not detected(need to improve this) > - creating descriptors is a bit slow(see TODO item in Descriptor.cpp > how to fix this)
Sory for the inconvinience, only the path to the library is wrong. I
will check if there is a better solution to this problem, meanwhile
you can fix this problem temporarily like this:
Replace my path to vigra with your path in this line(CMakeLists.txt):
TARGET_LINK_LIBRARIES(MatchPoint /usr/lib/libvigraimpex.so png jpeg
tiff)
So, /usr/lib/libvigraimpex.so should go to wherever vigra library is
located on your computer.
@Harry: yes, at the moment you still need autopano if you want to
detect control points needed for panos, because MatchPoint currrently
includes only feature detection but not feature matching.
cheers, Zoran
On 16 feb., 14:57, Yuval Levy <goo...@levy.ch> wrote:
> On Feb 16, 10:43 am, Zoran Mesec <zoran.me...@gmail.com> wrote:
> > Replace my path to vigra with your path in this line(CMakeLists.txt):
> > TARGET_LINK_LIBRARIES(MatchPoint /usr/lib/libvigraimpex.so png jpeg
> > tiff)
> no change for a few minutes. TOP still shows MatchPoint at 99%-100%
> CPU usage and about 200m memory usage.
> I'll let it run for a while.
> If you make changes and want other to test them, please take two
> minutes to update <http://wiki.panotools.org/ > Hugin_Compiling_Ubuntu#MatchPoint>
> I don't know when I will reboot into Windows to try it there. Don't
> want to interrupt ;-)
On Feb 16, 2:19 pm, Zoran Mesec <zoran.me...@gmail.com> wrote:
> Thanks for feedback
I'm back. we've spent the afternoon out in the snow. it was great.
hugin says elapsed time: 2:53:17 and a warning that the six images are
unconnected. those where my usual fisheye 8mm images...
> I suggest smaller images because there is no real code optimization
I like real life tests, and I don't mind the computer churning for
hours when I'm gone ;-)
You wrote: >Replace my path to vigra with your path in this line(CMakeLists.txt): >TARGET_LINK_LIBRARIES(MatchPoint /usr/lib/libvigraimpex.so png jpeg >tiff)
>So, /usr/lib/libvigraimpex.so should go to wherever vigra library is >located on your computer.
I did this apart from the fact that I specified .dylib instead of .so on OSX, but I keep on having the same errors. I also included the vigra source directory already in my path and I also tried with the static version (*.a).
Due to the work required for fixing bugs in hugin, I haven't had much time for testing MatchPoint in the last days. I just ran it on a simple 3 image pano and it worked fine.
Once I have tested it a bit more, we can probably iron out some more bugs and include it in hugin 0.7.0. This way, people in the US can be happy, too! (And a nicely working version of hugin will be in Fedora.)
stereo_sl wrote: > I have commited my latest changes for MatchPoint, suite for detection > of control points(GSoC remember?). The suite is now somewhat stable > and can be used as a replacement for generatekeys from existing > autopano application. I would like someone to give it a try, > preferably someone who has experience with existing autopano > application. In the svn folder there are also 2 images for testing > purposes.
> There is no gui, so those willing to test it, will have to: > - compile it using cmake& make(I can provide linux binaries on > request) > - invoke the app from the command line > - [optional] use -t argument to output file used for descriptor > performance testing with matlab(Mikolayczik suite)
Have you already done this? I have updated my linux system and need to reinstall matlab before trying. This test would give valuable information where further work might be required. I think this is more important than speeding it up.
I've been using autopano-sift to generate keypoints and do matching
and thought I'd give this a try. It compiled fine (I followed the
simple instructions for Ubuntu; I'm running Debian etch, 2.6.18) and I
got it to generate key files for me, but autopano doesn't find any
matching keypoints between the images -- and I know they exist since I
ran this test using generatekeys a month ago. On the plus side, it
works quickly. Here's the run for my six images:
============================================
Tue Feb 19 17:40:16 EST 2008
Height:2304,width:3072
Detected points:5888 Average:5697.02 Measured time:5.66
Creating descriptors...
5888
Generating XML file: /joe/originals/jr/105_0506.jpg.keys.xml
Tue Feb 19 17:40:52 EST 2008
Tue Feb 19 17:40:52 EST 2008
Height:2304,width:3072
Detected points:3218 Average:4245.54 Measured time:5.86
Creating descriptors...
3218
Generating XML file: /joe/originals/jr/105_0507.jpg.keys.xml
Tue Feb 19 17:41:15 EST 2008
Tue Feb 19 17:41:15 EST 2008
Height:2304,width:3072
Detected points:1622 Average:2833.54 Measured time:5.03
Creating descriptors...
1622
Generating XML file: /joe/originals/jr/105_0508.jpg.keys.xml
Tue Feb 19 17:41:30 EST 2008
Tue Feb 19 17:41:30 EST 2008
Height:2304,width:3072
Detected points:952 Average:1727.19 Measured time:5.82
Creating descriptors...
952
Generating XML file: /joe/originals/jr/105_0509.jpg.keys.xml
Tue Feb 19 17:41:43 EST 2008
Tue Feb 19 17:41:43 EST 2008
Height:2304,width:3072
Detected points:679 Average:1150.48 Measured time:5.76
Creating descriptors...
679
Generating XML file: /joe/originals/jr/105_0510.jpg.keys.xml
Tue Feb 19 17:41:53 EST 2008
Tue Feb 19 17:41:53 EST 2008
Height:2304,width:3072
Detected points:382 Average:605.367 Measured time:5.74
Creating descriptors...
382
Generating XML file: /joe/originals/jr/105_0511.jpg.keys.xml
Tue Feb 19 17:42:03 EST 2008
============================================
I also found that if I use --ransac off when running autopano I get a
System.NullReferenceException from BondBall.InitiateBond. Let me know
if you want the rest of it. I have a hunch that belongs in a
different thread (happy to start it -- but I haven't looked for info
about the problem yet.)
I'm anxious to try this -- I have 3,000 images that a friend & I took
in Greece, and 3/4ths of them are HDR or pano (or combination) input
pics. Anything that'll help me automate this would be terrific.
-j
On Feb 16, 8:16 am, stereo_sl <zoran.me...@gmail.com> wrote:
> I have commited my latest changes for MatchPoint, suite for detection
> of control points(GSoC remember?). The suite is now somewhat stable
> and can be used as a replacement for generatekeys from existing
> autopano application. I would like someone to give it a try,
> preferably someone who has experience with existing autopano
> application. In the svn folder there are also 2 images for testing
> purposes.
> There is no gui, so those willing to test it, will have to:
> - compile it using cmake& make(I can provide linux binaries on
> request)
> - invoke the app from the command line
> - [optional] use -t argument to output file used for descriptor
> performance testing with matlab(Mikolayczik suite)
> Usage:
> # first extract features from the first image and output them to the
> image1.key:
> MatchPoint image1.jpg image1.key
> # for second image:
> MatchPoint image2.jpg image2.key
> #match features from the two generated files using autopano:
> autopano project.pto image1.key image2.key
> # open the project file in hugin:
> hugin project.pto
> Notes:
> - points close to image edges are not detected(need to improve this)
> - creating descriptors is a bit slow(see TODO item in Descriptor.cpp
> how to fix this)
On Tue 19-Feb-2008 at 15:16 -0800, joemadeus wrote:
>I've been using autopano-sift to generate keypoints and do matching >and thought I'd give this a try. It compiled fine (I followed the >simple instructions for Ubuntu; I'm running Debian etch, 2.6.18) and I >got it to generate key files for me, but autopano doesn't find any >matching keypoints between the images
The simplest explanation is probably that MatchPoint currently only finds features in the central 70% of the images, so you need something like 50% overlap to have any chance of getting control points.
> The simplest explanation is probably that MatchPoint currently only
> finds features in the central 70% of the images, so you need
> something like 50% overlap to have any chance of getting control
> points.
Ok, but these images overlap 100% (well, 95-98% maybe.) They were
taken for HDR purposes, so they have nearly exactly the same scene,
just different exposures.
-j
On Feb 19, 7:17 pm, Bruno Postle <br...@postle.net> wrote:
> On Tue 19-Feb-2008 at 15:16 -0800, joemadeus wrote:
> >I've been using autopano-sift to generate keypoints and do matching
> >and thought I'd give this a try. It compiled fine (I followed the
> >simple instructions for Ubuntu; I'm running Debian etch, 2.6.18) and I
> >got it to generate key files for me, but autopano doesn't find any
> >matching keypoints between the images
> The simplest explanation is probably that MatchPoint currently only
> finds features in the central 70% of the images, so you need
> something like 50% overlap to have any chance of getting control
> points.
