View allAll Photos Tagged Hugin
Visited MOMA, had a lot to think about afterwards.
Like.. modern art vs. sci-fi art. Draw a circle and then make angular collections of rectangular prism forms above it... well, that's a space ship orbiting a planet. Or a piece of minimalist art. Being a closet minimalist, where you intend to make science fiction art or electronica music but instead wrap it in the jargon of fine art and sell it as minimalist art and music is a good way to make money and get invited into certain houses... whereas doing scifi art eventually devolves down into the lucrative furry marketplace.
Or of several modern art projects that I'd like to do but don't seem to get around to doing.
Or of that uncomfortable feeling, after seeing bunches of Edvard Munch art pieces in a gallery and realizing that I actually kinda understood him more than ever before.
Anyway, 8 photos handheld and stitched in Hugin
The algorithm for finding control points is extremely robust, but it can fail in two ways: it can fail to find a pair of points, and it can pair similar points found in the wrong locations.
I have noticed that it is especially prone to failure if the images were shot with different lenses and have different fields of view. That certainly affects the pixel structure of the control point areas, making it more difficult to match them.
In this case, the left frame was shot with the focal distance of 6.2 mm, and the right frame was shot at 8.6 mm. The point number 4 in this pair, although it shows very similar colours and is found in the same vicinity on both images, shows almost entirely dissimilar paterns and is probably bogus.
Another defect of the control point assignment shown here is that the points are clustered in a relatively small area. That is normally not a problem; if the frames are shot with the same calibrated lens without parallax, it is enough to have one or two control points to stitch the frames without any defects. But in this case, different fields of view and a serious parallax compound the problem.
Total of 6 images in this panorama, each at 28mm focal length, auto-stitched with Autopano-SIFT-c in Hugin. I re-applied the central image (as output by Hugin) as that was the sharpest one of the rather lovely organ pipes, and hand-blended the edges.
Horizontal field of view is 70 degrees.
Projection is Rectilinear, which gives those nice straight lines on the floor.
A panorama of the square off Gamsakhurdia street in Batumi. Needs to be cropped slightly to remove the blending artifact on the right side (I did the best I could with masks).
Previous: Any overlap is helpful
Once all pairs of overlapping frames that are assigned a reasonable set of control points, a single command ("Align" in Assistant) will send the job for alingment. Hugin manages all configuration data for a panorama in a text file, and it runs all jobs by invoking a rule or several in a Makefile that it creates to launch the external tools. It pops a window showing the log messages from the tools as they chug along. It is, an essence, a front end to make: it translates user input into a Makefile that is then use to run the tools, such as enblend and nona.
Once the alignment job is complete, a window like this will pop up. It shows a crude alignment of the remapped frames and provides a set of functions to manipulate the geometry of the panorama, such as projection and horizon-straightening.
There is a similar preview tool in Hugin called "Fast Preview Panorama". It has several more functions and uses a different input method for warping. Neither the "Preview Panorama", nor the "Fast Preview Panorama" can alter the alignment, but both tools can be used to selectively disable certain frames or change their bleding order. It is possible, for example, to disable a frame and then rerun the alignment of the remaining frames using the functions in the principal Hugin window.
Note that the alignment shown in the previews is approximate; even the images that look terribly misaligned in the preview will normally look great in the blended output image. Not in this case, though. The misalignment seen on the verge of the road will remain, because no algorithm can recover the information lost due to shooting from different points of view. In other words, if all subjects in the overlap area are visible in both frames, the frames can be warped to align. But if there is a substantial part of the scene that is visible in one frame and obscured in another — bad luck.
But I am not concerned with alignment at the moment. It is not too bad. The distant objects are aligned perfectly; the problems are concentrated in the front, where I can edit them away. My concern now is the exposure. Because all frames are exposed differently (I shot them in automatic mode, striving for the best fit of the histogram in each), Hugin optimised the exposure in each frame to provide for a smooth blend with the adjacent frames. As a result, the highlights in the rightmost frame are overblown, even though I anchored the exposure on this frame (Hugin allows to choose which frame to use as a reference for alignment, and which frame to use as a reference for exposure optimisation).
There is a simple solution to that. The overall exposure value can be adjusted to a desired level without the loss of information (or so it seems).
My first Pano! Used Hugin to stitch 2 blended expsoures. These were derived from 5 bracketed shots each that were blended with Enfuse. The resultant image was fine tuned with CaptureNX2.
Previous: Adjusting the gamma curve
Building the frames for Hugin
Once I have reviewed all frames and added the build rules to my Makefule, I add this rule:
AonachDubh.pto: aonach_dubh-f1.tiff aonach_dubh-f2.tiff aonach_dubh-f3.tiff aonach_dubh-f4.tiff aonach_dubh-f5.tiff aonach_dubh-f6.tiff
echo "create AonachDubh.pto with Hugin"
In the absence AonachDubh.pto, this will work as a dummy target, causing make to generate the TIFF versions of all frames. Once I have created the panorama file AonachDubh.pto with Hugin, an as long as all frames are present, this rule will have no effect.
Automatic alginment
Adding the frame files to a new panorama in Hugin using its Assistant interface will make it assign the control points and attempt to align the images in one go. With a good enough set of frames, such alignment will normally run without problems, but with this odd set I've got the first result was a bit disappointing.
The mess shown here is caused by a combination of factors, including the lack of overlap between some frames, bogus control points found in the farmes that did not overlap, too few control points between well-overlapping frames, too many control points concentrated in very small areas, and perhaps other problems I am not aware of. The solution in this case was to delete the automatically assigned control points and place them manually, aiming for the best spread.
But before we do that, let's take a look at a few bad cases of control point positioning.
Spherical map of a panorama I took in Tarragona's roman amphitheatre.
Shoved together with Hugin, and mapped into a planet. First time ever for trying this.
Antes de sair o sol, comezamos o descenso ao Gran Canón do Colorado (USA) polo camiño Kaibab Trail:
deviaxe.blogspot.com.es/2012/07/o-gran-canon-do-colorado....
Used Hugin software to stitch 8 snaps from my 70d on the road from Santa Maria Degli Angeli to Assisi.
Best seen at full resolution.
lujiazui shanghai paltform suway station
hugin panorama group (china) at:
www.flickr.com/groups/love_hugin
Let's join us!