my work around has been to just stich them as two seperate images, and pull them together in gimp/PS during post processing. Ive had mixed results depending on how close exposure/colours etc are between the two sets

If you "de-fish" the fisheye lenses, can APP use that?

Yes, if you defish the picture and then use it, it will work. The multifocal support in APP is really working great in v2. I made some 4 Gigapixels panorama with this technic ( 70mm for sky and 200mm for ground, it just awesome ).

I've noticed that image sharpness matters, blurry parts of images don't always stitch to sharp parts as well as sharp parts of images stitch to sharp parts of other images.

So it may not be a simple matter of a focal length difference.

Alexandre,

I usually use PS on the output PSB/PSD files, but there is a color difference at the edges of pictures. 

Your technique allows the blender to do its job.

But your technique also is against the APP philosophy of "everything automatic as possible".

One way to do this is to allow us to paint transparacy in APP on each image. 

I also want the ability to paint transparency on the pano editing window to paint on the sky, and on waves, and moving objects.  This would say "don't use these CPs." 

But your way (normal and telephoto shots) is more interesting.  If we could paint a transparancy over part of the image in the pano editor, and then manually select the image parts to show there, it may be more effective.

I envision:
1. paint a region transparent.  APP puts a mask over this region (or over the rest of the pano).  You could paint the mask on or off.
2. click within the region.  The mouse wheel would allow you to scroll through all the underlying images (as the CP editor does).  Each image would be defined (eg use a border, highlight that image,etc), so you could see how much of the region it covers.
3. Right click would give an option to use the image.
4. several images may be selected within a region.

CPs within the region would not be used, except possibly CPs on overlapping selected images that are used.

Last edited by hankkarl (2009-05-29 17:31:16)

AlexandreJ wrote:

I agree with you hankkarl. We are working on a blender that will use this focal length different to first keep the one from the longest focal.
The solution for the moment is either render several layer, one per focal then, blend manually in photoshop, or paint transparency. I did that once for a gigapixels :
70mm for the sky and 200mm everywhere else. The order of operation is important :

1. Detect the panorama ( jpeg in entry for example )
2. Convert the 70mm into png format, paint the ground transparent so the ground is forced to come from 200mm images
3. Replace in the .pano file the .jpg by .png for the image you edit
4. Render

It works really well ( if you don't forget to shoot some zone at 200 mm, I did this mistake ).

Just thought--can you start with PNG, then edit PNG for transparance, then render?  you don't have to edit the pano file if you can do that.

AlexandreJ wrote:

Yes I agree, but be aware that their is 2 kinds of mask at least :
- mask that says, don't put control point here
- mask that says, don't use this pixel for the blending

My first guess is that the mask that says "Don't put control point here" could be done in the pano editor on all pictures for sky and waves, but has to be done on each picture for tourist removal. 

And the mask that says "don't use this point for blending" must be done on each picture. 

It may be good to have a mask that says "prefer this pixel-but don't use it for blending" for tourist insertion and selecting one image of the sky, waves, etc.

Just to say i have had great success using the method outlined by alex in post #9 of this thread, see this thread in the gallery for a little more info: http://www.autopano.net/forum/t6498-gigapixel,of-death

HansKeesom wrote:

At the moment I thinking about and investigating another approach. I would like to know if other have experiences with it.

I make the photos for a 360*180 pano using a 10 mm fisheye. Horizontal rotation of 60 degrees.
After making the needed photos (2 rows plus two shots down) I change the lens to a 50 mm lens and make one round using 10 degrees steps (36 photos).
I stitch the first photos together into a 360 *180  pano.
I then load the 36 other photos and add the pano I just generates. I tell autopano to include all photos in the pano.
Autopano then places the 36 photos on top of the pano and does so rather well.


Like to hear some comments on this

So how do you have a way to mask out the parts made with the 10 mm fisheye or do you assume/know that APG uses the file with the most data preferentially or even averages them? From what Alexandre has already said previously, one shouldn't be able to stitch fisheye with rectilinear images in one pano!

But you say it works! What do you think is happening to the detail? If you stitch just the 50mm images in a separate pano, do you see the same detail in these areas or is it superior to the one blended with the 10mm fisheyes stitched underlay of images?

Asher