My Autopano Pro forum

Certainly a fine illustration of how APP works but not a fine illustration of the way this can lead to beautiful images!

Here is one of the used shots:

Last edited by GURL (2007-01-24 10:09:47)

George is right, one of the coolest ways to use APP is stitching a mix of macro images and more distant ones.

How did you decide where to set focus for each shot and why did you settle on f/8 instead of f/11 or f/16? I assume that the way this works, APP is replacing the out of focus area of one shot with the in-focus area of another?

Thanks GURL another very nice contribution.

Those who find this interesting might check out norweegian photographer Mikkel McAlinden: http://www.mikkelmcalinden.com/

I made another attempt where the subject is about 10 feet (3 m) from the camera and about 60 cm (2 feet) deep (from the cat nose to end of the cat tail.)
Focal length (35mm equivalent) was 240 mm - aperture was f/8 - depth of field (as found in the EXIF data - I recently discovered this useful info being hidden among many useless or useful ones in EXIF) was 12 cm (4.7").

I stitched 12 shots and found Multiband giving much better results than Smartblend (some improvements of Multiband result are possible using a PSD layered file but differences need an attentive examination.) I can't compare with Helicon Focus but suppose some APP users could be interested in using APP for that kind of images.

Here are some very interesting tutorials or discutions about DOF, COC, diffraction, etc:
DOF: http://www.cambridgeincolour.com/tutori … -field.htm
Diffraction: http://www.cambridgeincolour.com/tutori … graphy.htm
How to take identical images using a compact digital camera and an advanced DSLR (when possible, that is): http://www.luminous-landscape.com/essay … nses.shtml
Résolution, MP, profondeur de champ, décalage MaP (French): http://www.photim.net/nci/discu.php3?co … lmimage#O0
... where I learnt that DOF is independant of the used lens (but is much less visible when a wide-angle lens is used than when a long lens - which enlarge it - is used), that for a given sensor the F stop where the detrimental effect of diffraction begins don't depends of focal length (provided the lens is good enough), that a Canon G7 camera should always being used at F/2 to actually make full use of its 10 MP (but somewhat down-sized) sensor and more.

GURL  wrote:

An interesting point is that SIFT is able to place CP on features wich are crisp in one source image but are fuzzy in the other one:

Those CPs look close but not exactly right.

#67 looks to be a couple of pixels lower in the right image than in the left (not so bad),
but #52 is noticeably further right, and maybe a little lower, in the right image.
Was the alignment error in the final stitch about the same size as these CP errors?

I'm a new user of APP, and I'm generally very satistifed with the results.  I have noticed that
SIFT has trouble when trying to match an out of focus image to a sharp one.

Hey guys,

I'm trying to reproduce those results, but I must be missing something... For example, a simple in-doors shoot of my desk, 5 100% overlapping photos, taken from a tripod, variable FL. I drop those in APP, detect with no lens correction, choose planar projection, render as Multiband TIFF and the result is no good - parts of the images are mixing in a weird out of focus/in focus mix.

Of course, rendering as PSD with layers and selective deletion of image parts works, but it takes time. Somehow I got the impression Georges got this working in a single click manner?

There is alghoritm to make DEP blending

1. You need 'smartblend' with support of alpha channel (you plan to make this kind of blender in future)
2. You need to make alghoritm to analyze sharpness of small part of image. This algoritm will build sharpness alhpa channel for each image (additional channel to source alpha channel). Transparency of pixels in this mask will be proportionate to sharpeness of source pixels
3. 'smartblend' will use this mask to build pano with maximum DEP

Similar method can be used for HDRI-smartblend, but with another alhpa cnahhel - quality of pixels's color (pixels value near to 0 and 255 have minimum quality, in range 64-192 - maximum quality)


This is only abstract idea to think

Chemop wrote:

How about this one, taken last week in the SW Australia Karri forest!

Showing very close items and very distant ones in the same image is not something usual, neither in painting nor in photography. I feel like having a new tool and now what? so that I like viewing examples of very large DOF images like your's...

bo wrote:

Somehow I got the impression Georges got this working in a single click manner?

For the two examples above I made two versions:
1) a first one (from uncorrected JPEGs) where I let APP minds its own business: the result was not perfect but was much better than when using a single focusing distance
2) a second one (from RAW processed and distortion corrected TIFFs) where I used the links Editor to remove CPs (some CPs were set on the pano-head in the tools example - a very large amount of links was created by APP in the cat example so that suppressing the worst ones helped a lot.) Resulting file was a multi-layer PSD including the multiband stitched pano and each of the aligned source images.

The RMS difference between "auto" version and "manual" versions  is large (from 4 to 1.3 for the "lady on the cat" example (where, probably, too many shots were used.) But, whichever the RMS, I insist that APP "automatic results" are much better  than any of the single shots I did. The higher the distance differences are in the subject the more the automatic result looks useful.

Last edited by GURL (2007-02-05 13:15:47)

To [bo] : what I wrote was obviously written too fast : right sentence here :

"Yes, he managed it because his shoots are not 100% overlapping. So APP made a smooth blend between them achieve by the same way a better depth of field. If all pictures are just stacked upon each other, it can't do it."

Hmm, I did not realize that Georges' photos are not overlapping... I will try again with similar approach.

Just a follow up. I inspected closely your last post with the cat in the weekend, when I had time to experiment.

I found that actually your result is as flawed as mine. There are sharp contrast between focused and blurred areas without any transition between them. So, I have not done anything wrong the first time, it's just a matter of luck and even then you get not nearly perfect results.

Even with completely overlapping shots, it is possible to use APP, but you still have to mask and delete layers (partially) in Photoshop.