 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | You are not logged in.
Pages: 1
AutoPano Giga 2.0.X fails to optimise the X and Y offset of the fisheye lens relative to the sensor correctly. This is a vital adjustment for fisheye panoramas - without it the distortion is adjusted incorrectly and it's impossible to get a good stitch.
I had almost given up on version 2 until I noticed that it wasn't adjusting the offset correctly during optimisation. By looking back at my old .pano files from version 1.4.2 in a text editor, I was able to calculate an average value and input this manually. Suddenly, panoramas that stubbornly refused to stitch well in v2 just fell into place.
Here are the offset values as calculated by v2's optimiser:
<FisheyeCoffX> 0.00778917 </FisheyeCoffX>
<FisheyeCoffY> -0.00324429 </FisheyeCoffY>
And here they are as adjusted by the optimiser from my manually set values (-22 and 20):
<FisheyeCoffX> -21.9961 </FisheyeCoffX>
<FisheyeCoffY> 19.9995 </FisheyeCoffY>
Strangely, the difference it makes in the RMS is pretty small - 2.7 - 4 without setting a value manually, 2.0 - 2.3 if it is set. But the visual difference is striking. See the attached images for an example.
If I hadn't previously owned v1.4.2, I wouldn't have been able to work out why there was a problem and work around it. The quality of the stitch was so disappointing I'd just have abandoned v2 altogether. I hope this can be fixed.
Last edited by Corvus (2009-08-05 16:38:57)
Offline
Corvus wrote:
See the attached images for an example.
please resize it a lil bit to more than 1x1 pixel 
Offline
Corvus wrote:
Strangely, the difference it makes in the RMS is pretty small - 2.7 - 4 without setting a value manually, 2.0 - 2.3 if it is set. But the visual difference is striking. See the attached images for an example.
The RMS of the link corresponding to the attached example seems larger than 10. A different placement of control points by the new version could be enough to cause such a bad link.
Does the project produced using version 2.0 results in a bad RMS when it is optimized (again) using version 2.3?
Offline
GURL wrote:
Corvus wrote:
Strangely, the difference it makes in the RMS is pretty small - 2.7 - 4 without setting a value manually, 2.0 - 2.3 if it is set. But the visual difference is striking. See the attached images for an example.
The RMS of the link corresponding to the attached example seems larger than 10. A different placement of control points by the new version could be enough to cause such a bad link.
Does the project produced using version 2.0 results in a bad RMS when it is optimized (again) using version 2.3?
Sorry, I think I may not have expressed myself clearly enough. The range of RMS values I get without manually setting the offsets is between about 2.7 and 4. The range of RMS values if I do set the offset is between 2.0 and 2.3. However, the remaining distortion in the images is enough to leave large stitching errors like the one in the sample image.
Offline
Seems you found a real confirmed issue => I'm studying this right now.
Offline
Confirmed issue => solved. The quality is not a bit better with camera/lens which are strongly off centered.
Offline
AlexandreJ wrote:
Confirmed issue => solved. The quality is not a bit better with camera/lens which are strongly off centered.
I am not clear what you mean, Alexandre, when you say solved.
Will this be in the next bug fix release?
I have experienced similar problem with fisheye stitching in AGP and would like to see this resolved.
Thanks
Brian
Last edited by BrianLR (2009-08-27 02:34:31)
Offline
Yes, when I say fixed, means that I found the reason here and that I fixed the bug in our codebase.
So the next release will have this issue resolved ( in v2.0.4 which is of course not yet available ).
Offline
Hi Alexandre,
Will this fix also apply to rectilinear lenses? I'm thinking of my 12mm on FF -- would it have similar centering issues?
Yes, it should be rectilinear, but it does have distortion. And APG "distorts" images into almost a circle to fit them into the pano, so centering may matter here.
the more common 16mm (eg 10mm on 1.6 crop) may have the same issue to a lesser degree.
Offline
AlexandreJ wrote:
Yes, when I say fixed, means that I found the reason here and that I fixed the bug in our codebase.
So the next release will have this issue resolved ( in v2.0.4 which is of course not yet available ).
Can you give any indication of how long before v2.0.4 will be released, is it days, weeks or months?
Offline
We don't do offset calculation for rectilinear lenses. I had that study on the todo but didn't found yet really off centered standard lenses.
It's weeks ( perhaps end of next week, 4 september but more probably the 11 september ). Don't take this date for anything sure. If we found something that we need to solve, we'll postpone.
Offline
AlexandreJ wrote:
Yes, when I say fixed, means that I found the reason here and that I fixed the bug in our codebase.
So the next release will have this issue resolved ( in v2.0.4 which is of course not yet available ).
Wonderful, many thanks! It'll save a lot of time...
Offline
Pages: 1