I try to model a room using pano's from Stitcher. The pano's consist of pictures taken in the corners of the room. I provided Stitcher with the right film back and focal lenght. The planar renders (about 3000x2000 pixels) seem perfect (aside from a few blurred edges here and there). Cam data is exported and assigned to each pano in IM. After placing 40 locators i get no automatic calibration, forcing it gives only red icons... Is it possible that the Stitcher pano's -although 'visually' perfect- have slight deformations wich make them unreliable for good calibration in IM? Included a small version of one of the pano's. Thanks for any suggestions.

That's so funny -- I was just thinking about trying exactly the same process earlier this morning, and had come to my own conclusion that it would be best to avoid it. Here's why:
 
My guess is that IM cannot ascertain the "lens" properties from the planar renders coming out of Stitcher. IM must must have these lens properties to understand the "form" of these objects in 3D.
 
Based on my (much more extensive) experience with Stitcher -- AFAIK, the same engine is used to calibrate the lens -- differently sized, oriented, or processed images for input will confuse the calibration engine.
 
Also, every time you have to distort or remap the projection of an image, you are going to get blurring or otherwise loss of information that cannot be reconstructed -- sort of like making a photocopy of a photocopy. So, considering this, you lose something -- though slight -- going from images to planar (in Stitcher), and then you lose something again when going from planar to the UV-mapped texture (in IM).
 
Try instead inputting unstitched images. This would save you a step, and probably produce higher quality results in the end anyway.
 
Hope this helps.

Thanks, I'll give it a try with single images. The workflow with the Stitcher's planars seems to be a common worklow though : the church example is constructed that way, it's a pity there's no further information on the do's and don'ts of such workflow.

Actually, right after I posted my response, I read a more precise take on a similar situation:
 
http://www.realviz.com/board/viewtopic.php?t=876
 
Apparently, if you stitch more than 3 or 4 images together, you are pushing the FOV wider than the calibration engine can handle.
 
But even if you back off to 3 images, you still are dealing with TWO generations of resampling on the textures, though, so that's something to consider if you're trying to retain the highest possible quality.
 
IF the IM algorithm for interpolation/mipmap-generation is the same as it is for Stitcher, then this would NOT be a small compromise in quality to resample twice. I've exaustively tested the interpolation algorithm in Stitcher and posted my results here:
 
http://www.carbonltd.com/interpolation/
 
(Fortunately, since I've written this, the problem has been acknowledged and is being worked on.)
 
Can RealVIZ please confirm whether the interpolation/mipmap algorithm in ImageModeler is the same as in Stitcher? It would be nice to know, so that I can tweak my production process for best results based on that knowledge.
 
Thanks mucho in advance...

Hello,
 
  The interpolation in IM is the same as in Stitcher. But the results can be different, as:  
- IM does blending using several images, so a textured pixel is a certain sum of pixels coming from several images.
- Due to the perspective of the elements in your image, the plane to texture is rarely parallel to the image plane (in space), so the textured image gets some distortion.
 
 
Regarding the question "using panos or not":
 
For sure, if you can avoid using panoramas, you'll get better quality, and precision. This can be done generally when the object or the scene to reconstruct can be captured with "classic" lenses, i.e. objects that you can shot moving all around (statue, house, little objects, etc...).
 
But for interiors or places in a town (see Nice plazza on www.gregdowning.com) using stitched images will help a lot for the calibration process.
 
If you need higher precision on some places one could mix stills with panoramas...
 
Hope it helps,
 
Stef

 
That's what I suspected. So with the new version of Stitcher supporting better interpolation methods (like the Lanczos option I had really pushed for ), will this enhancement make its way over to ImageModeler?
 

 
Does the IM calibration process aim to describe the lens alone, or is it also trying to establish a starting point with which to describe the scene in 3D space? (For that matter, would a Stitcher lens profile help at all as a starting point for ImageModeler -- presuming they refer to the exact same lens -- or are these completely different animals?)
 
I'm about to dive into a project which would require several interiors, and I'm hoping to understand what precisely about this situation ImageModeler benefits from:
- if I'm able to provide enough images containing enough features (in disparate parts of the image, lying on different planes, of course), would this alone be enough for calibration?
- or does IM require the pano for it to understand the "big picture" of how these complete shapes fit together -- on a macro level -- as they might span beyond the frame of most single images?
 
It would be really helpful for me to understand this before I waste time on it.
 
Thanks mucho in advance.

Hi,
 
About interpolation:
Yes, this is a enhancement we can add in Imagemodeler. It could be an option at the texture extraction step (like in Stitcher). It will slow down the process a little bit.
 
About interiors and panos:
 
Calibration: the internal parameters (focal length, distortion) and the external parameters (position, orientation) are estimated in the same process at the same time. They are not separated steps in IM ...  
But some of them, focal length and distortion, can be set as known (value known, will not be estimated), constant (value constant for all shots, ex: focal constant) or variable. For constant and variable settings, the value set is used as starting point in the calibration process.
 
Checking "save camera" in Stitcher rendering dialog creates a .cam file that you can reload in IM. Giving known values to IM simplifies the calibration, but you have to be sure of those values.
That's why I suggested in an older post:
- take 3/4 for picture in a row, and 2 rows minimum
- about 40% overlap
to have a good estimation of the focal length in Stitcher.
As you may know, the only way to be sure of the estimated focal length is to close the panorama... But my experience showed me that 3x2 or 3x3 panoramas are accurate enough.
 
Some tips for interior projects:
- try to have features seen in at least 3 shots (that's why panoramas are helpfull)  
- add an object or more in the center of your room; it gives you features seen in all shots
 
Another solution might be to use Sceneweaver that is able to calibrate one unique panoramas of a room (SW needs a corner in the panorama, to be able to calibrate it). This is a good solution if your room is not  too complex, i.e. few objects are inside the room.
 
Hope it helps,
 
Stef