 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | Rules | You are not logged in.
Hello Everyone,
The screenshot below shows the result of importing a 700-image gigapan into APG2. As you can see it has collapsed and deformed the sky. Interestingly, the way it has also deformed the London Eye is similar to the mess that the Gigapan stitcher made of it (see here).
That is not the main bug however. The real problem with this pano is that it has made a 360 out of a 115-degree pano. If you look at the second screenshot (compare it with the gigapan referenced above) you can see that the first column of images is in the middle of the stitch.
Do you want the files Alexandre? I'll have to post them.
Aeris
Last edited by Aeriscera (2009-05-13 21:54:13)
Offline
Ok, my guess about this case : autopano found a wrong link between some building from the first column and the last column.
It forced the panorama to wrap but should not ( can you look at that with the CP editor ? )
Second guess : there might have been a lot of moving cloud this day, right ?
Offline
This pano is a great example showing that control points are often to be avoided when stitching the many images in a gigapixel mosaic !
This suggest a new feature would be usefull to avoid Autopano trying to set CP in the blue sky, on the moving clouds, on the rotating (?) London Eye wheel and even on the quiet Thames water...
Offline
GURL wrote:
This pano is a great example showing that control points are often to be avoided when stitching the many images in a gigapixel mosaic !
This suggest a new feature would be usefull to avoid Autopano trying to set CP in the blue sky, on the moving clouds, on the rotating (?) London Eye wheel and even on the quiet Thames water...
But Gigapan Import filter surely 'knows' (or could 'know) whether a pano can/could be 360 FOV horizontally from number of columns and image FOV? And hence could be able to prevent some spurious links such as some of those shown in the example?
Last edited by mediavets (2009-05-14 11:20:59)
Offline
AlexandreJ wrote:
Ok, my guess about this case : autopano found a wrong link between some building from the first column and the last column.
It forced the panorama to wrap but should not ( can you look at that with the CP editor ? )
Second guess : there might have been a lot of moving cloud this day, right ?
Your first guess is right, see screenshot. I went into the editor to delete the links between the first and last columns in the hope that it would unwrap after optimisation. Unfortunately it is not just the first and last columns that are connected - it is more like the first and last five. Worse, I get the GFL bug I raised the other day every 10 links, so removing all the incorrect links is not really practical. Unless you have a workaround?
Your second guess is probably correct too, but I am not sure.
Finally, you didn't comment on the way the top of the of the pano is not straight. The fact that the pano was taken with a motorized panohead suggests to me that it is wrong for the images to placed in anything other than a rectilinear grid. Of course there will be some adjustment - I am not saying the grid should be mathematically perfect - but I wonder if your algorithm places too much emphasis on feature-matching and not enough on the fact that a motorized panohead was used? In any case, I'd call this a bug. Do you think so too?
Aeris
Last edited by Aeriscera (2009-05-14 12:23:33)
Offline
Aeris,
Do Giagapan robot users ever use their robots to shoot panos with a 360 HFOV - can the Gigapan stitcher and viewer handle 360 HFOV panos?
If not it would seem to make sense to limit potential links to immediately adjacent images and if the Gigapan Import filter took into consideration HFOV being less than 360 then the sort of spurious links show in your example should then never arise?
Last edited by mediavets (2009-05-14 12:52:23)
Offline
mediavets wrote:
Aeris,
Do Giagapan robot users ever use their robots to shoot panos with a 360 HFOV
? Yes, all the time.
mediavets wrote:
can the Gigapan stitcher and viewer handle 360 HFOV panos?
Yes in principle, sometimes in practice - there is a known bug. Not sure what you mean by the "viewer". Gigapans can be viewed inside the stitcher which can handle 360º, the viewer on the web page doesn't attempt to, but the gigapan layer in Google Earth does and can.
A
Offline
There should be some workaround for this case by removing manually the false link and reoptimizing. But I cannot test that myself as I don't have any gigapan with such issue. If you share this project's image, I'll check them. BTW : you have our ftp for that purpose.
About 360 panorama. I could add a flag that will check for the first / last column. So for this case, just uncheck it and you won't get those link.
But ( there is always a but ... ) I'm sure it's not the only issue because it's not possible to get link between column 1 and 3 for example. The only possible links is column 1 and last column, but not last-1 for example. So if you have such links, that's a bug.
Offline
AlexandreJ wrote:
There should be some workaround for this case by removing manually the false link and reoptimizing.
As I said above this is not practical. I was thinking you might suggest editing the .pano.
AlexandreJ wrote:
If you share this project's image, I'll check them. BTW : you have our ftp for that purpose.
Again as I mentioned before, I can't afford to stuff up my internet connection for the time that is needed for large panos. This why I posted you a disc a couple of weeks ago. (You never acknowledged the disc - did you get it?) This one is 750 8Mpix JPEGs and my upload speed is at best 2Mbit/s. I may try it overnight tonight.
AlexandreJ wrote:
I'm sure it's not the only issue because it's not possible to get link between column 1 and 3 for example. The only possible links is column 1 and last column, but not last-1 for example. So if you have such links, that's a bug.
Yep see above. I have links between column N-3 and 1. (In fact between column N-3-i and column 1+i for i = 0, ..., 2). Initially I thought this was to do with too much overlap between images, but the rest of the pano looks OK (i.e. links between adjacent images only).
Here's a better view of the offending area showing a link between column N-3 and column 1:
Last edited by Aeriscera (2009-05-14 14:51:11)
Offline
