pollarda wrote:

I have just come back from a trip to California where I took some panos at Big Sur and Point Lobos.  When shooting the photos, I bracketed but when I processed them, I simply converted my regularly exposed "raw" image to a set of five images with exposures of : -2, -1, 0, +1, +2  (I figure I can get away with this a bit since I'm shooting with a D600 and a D800 both of which have plenty of dynamic range though I'm still curious as to how much I'm giving up by "cheating" in this way.)

Anyway, since the five images I use when generating each "stack" for the panorama came from the same exposure, they are identical in terms of content.  Yet when I stitch the images together, I have portions of the image which exhibit ghosting.  Why would the system align up most of the images in a stack properly and yet miss an image (or two) so that it doesn't align with the rest of the images in its stack?

Is there a way to avoid this?

I've tried this with the "No Ghosting", "Fusion" and "HDR" settings....  (Even so, the images should all align within a stack so these shouldn't matter much if at all.)

Thanks,

-Art

I tried your technique of using the same image processed 5 times to simulate bracketing and got exactly the same problem as you reported - that is the presence of ghosting across the overlap areas. When processing the same panorama just using the normally exposed (0) images then the ghosting was not there even if Exposure Fusion was nominally turned on

I tried all the same settings as you with the same results except I tried one other one and for reasons I cannot begin the explain it seemed to cure the problem - that was that I moved the slider on the preview screen in edit mode from priority to Ghosts along to priority to Long Focal (i.e to the extreme right hand side). This is the opposite of what would be intuitively expected but definitely mitigated the effect. The other settings I used were "Hard Links" + "Reference Level" in the detection stage and then just the defaults on the Fusion and Edit stages except for the slider I just mentioned

This only works where you have 5 exposures of the same image and not if processing real bracketed or stacked images taken individually in APG (which I have now abandoned in favour of Photomatix)

As for your other question about this method as opposed to real image bracketing - I wouldn't call it cheating but I would think the results can only be the same or worse than proper AEB especially on HDR where there really is a high dynamic range. If you cannot spot the difference on a parallel run of the two techniques in the final output then I guess it must be a good idea !

best
Martin

pollarda wrote:

Thanks for the help / input. 

Yes, it is strange that if the panorama is stitched with just the one layer (normal exposure), it processes just fine without ghosting. Yet when the same image is used (since it is raw and has the latitude) to create a series of pictures of varying exposures (-2, -1, 0, +1, +2) the ghosting appears.  The only thing I can figure out is that perhaps one of the layers in the exposure stack is getting "forgotten" somehow when it comes to doing the alignment since all photos in a stack should be aligned and transformed exactly the same..

I should point out that I imported the photos using the Gigapan importer. (I should point out that is vastly improved in this latest version -- though it still needs a bit of work IMHO.)  Even so, this does seem to help let Autopano know where things are at.

If I do a
Settings->Links->"Detect links in: The reference level or All stack levels
and select "All Stack Levels" as suggested it definitely helps.  I haven't tried the other suggestions yet.

Thanks again

-Art

I had another look at this just using 2 layers of 2 images in each layer and this ghosting problem is still there (sometimes). It never happens when just a single layer is stitched so it has got something to do with having > 1 layer and/or using the Exposure Fusion tools

The images used in each layer were (as before) identical images just processed differently via software to replicate the effect of bracketing so things like tripod movement etc can be eliminated

With 2 layers sometimes the ghosting was resolved and sometimes it wasn't. I couldn't see a definite pattern but the closer the two layers were to each other in terms of exposure the less likely ghosting was found. Layers of (0,0) (-1,0) (+1,0) were often OK but (-2,0) (+2,0) (+2,-2) (+1,-1) usually had ghosts present after rendering

I took a video in preview mode whilst using Exposure Fusion when stitching 2 layers of 2 images just to show how random this effect seems to be. Just altering the magnification of the preview screen or moving the slider from priority=Ghosts to priority=Long Focal can ignore or remove the ghosting in a fairly arbitrary way.

This is my previously published "moving ducks" panorama so you have to keep your eyes on the duck in the foreground

http://www.youtube.com/watch?v=O2Z_ZgT8Klk

Martin

Hi Georg

No just straightforward File | Select of 2 images x 2 to give a total of 4 images in 2 stacks of 2. In this case each layer is identical in composition (as it is the same shot N times) to the other layers except for exposure.

The duck is in the overlap area between the pairs of images and turned round in between shots so it is facing the other way round in the second shot of each pair

This is to me just a straightforward ghosting situation like you would get in a landscape shot with people walking around between shots in a panorama and should get resolved by APG