Sharing knowledge around Autopano
You are not logged in.
I have tried several times to add control points to an image pair, unsuccessfully. Where the previous versions added the new points quickly and accurately the new build takes forever and the adds the points accurately in one image and wildly outside the selected area on the other image.... This was with Canon 15mm fisheye.
That's if your lucky. App usually crashes in the process of trying to add control points to my canon15mm fisheye images. I just tryed again with defished images and had a crash during my attempt to add control points. I noticed this particular crash occurred when I moved the mouse during the several minutes while the pizza wheel was spinning....
I also tried the CP Editor with non-fisheye 24mm lens and all was normal, so evidently something about the fisheye is not alowing the CP editor to work normally.
Separate issue: I noticed when I rendered a bracketed pano in the new 1.4 A2 that the layerers do not line up...? They were OK in most of the pano but on the left the alignment was off by about 5 pixels. Also the minus 2 bracket had blank black areas in portions of the image...?
Last edited by Rico (2007-07-11 23:14:46)
Offline
Rico, I'm using the Win version of 1.4 and there is definitely a bug in the control points editor when working with full frame fisheye shots.
If Autopano gets the stitch right the first time, the spheres are fine. But if it misses - that's it, there's no way to add control points.
I have not had any program crashes, but obviously this bug need to be squashed.
I don't think the control point addition logic works in fisheye shots because of the completely different distortions between the two images. So you select a square that contains the same features but the general SHAPE of the features is so different that the program just can't find matching points. Your eye sees the similar points in the two images but the computer can't.
I think the only way to solve this is to include TWO ways of adding control points - the existing one for non-fisheye shots and the ability to add a particular point (as in PTGUI) and match that point between the two images.
Until this gets fixed, AP isn't a viable choice for professional sphere photography.
However, that being said, if you want to make spheres using non-fisheye images AP is the best there is.
Offline
In fact, control point adding in fisheye picture is far longer that for standard images. So it can take 10, 15s for some picture and I missed to add a visual progress bar for that. It's currently beeing done.
Offline
I've noticed that APP can't add control points in some cases. One example is when I used an 18mm lens (Sigma 18-200) on a 300D and used about f/11. Some parts of the picture were slightly out of focus, some were not. Edges were more distorted and the focus was softer than in the center. APP could not add points even where there were notable similarities (e.g. a metal plate with eight hexagonal bolt heads, letters, etc) if that feature was in the center of one image and the edge of another.
Offline
10 or 15 seconds would be no problem, however the wait is on the order of 5 or more minutes and often app hangs endlessly with no way to abort, if I move the mouse often I get a crash. Also the features I select are major features app should be able to detect IMHO. Progress bar would be good however I have tried this numerous times and there is definately a bug!
I finally got a very good render after several crashes during attempts to add CPs. I avoided the cp editor finally!
Last edited by Rico (2007-07-12 17:47:52)
Offline
Well, I wish Alexandre would take another look at this bug.... I just tried one more time and got the pizza wheel and no control points. I had to force quit as App was not responding. This is very frustrating as I went to great trouble to aquire these new images. Again it only happens with my 15mm fisheye, full frame canon lens. Thanks for your comments.
Offline
Seems like there are problems with manual cps on win too. EOS 300/Sigma 8mm....
Offline
Here's the issue. I'm pretty sure it happens when trying to stitch a zenith or nadir with another picture. I have to do some more checking with the picture because the fisheye search algorithm only work between picture rotated with the same angle. If one is +90° and the other 0°, it doesn't find any cp. I will improve this part to so it will work in either case.
Offline
In my case the problem was nadir shots, thats correct.
Offline
Alexandre, I did another test. I will give you a screen shot of the test. I tried to add control points to 2 vertical 15mm Canon fisheye images. It took several minutes for the spinning pizza wheel to stop and then added cp's on the left image correctly inside the selection box. (highest points on the horizon feature) In the right image the corresponding numbered cp's were placed wildly outside my selection box with no logic whatsoever.... So the theory that the bug only appears with nadir or zenith and another image at 90 degrees is incorrect....
This is a 360 QTVR but here also is a regular pano I extracted from the full sphere. Thank you!
Image Copyright 2007, Richard Sisk
Last edited by Rico (2007-07-17 03:39:55)
Offline
I can confirm this behaviour on win too....
Offline
[Win XP, 1.4A2]
I have the same problem for the CP outisde of aera or inaccurate.
I shoot with a D80 (crop 1.525) and a Sigma 10-20mm@10.
I dont have screen capture at this time, but i it need, i can do it !
Regards,
Arnaud.
Offline
Can someone of this thread provide me some pictures to test. I need some testing cases.
Offline
Alexandre, I think I stumbled upon the problem with adding the control points... I was preparing to send you some files to test and figured I better do 1 more test. I chose different files from those I had the problem with and found that the control points added correctly. Same lens(Canon 15mm full frame fisheye)
I then realized the only difference in the two sets of images is that I had bracketed the first set and then selected only the normal exposure set out of the bracketed series. So the images loaded into APP had image numbers separated by 2, due to the bracketing.
See the 2 images in the above example, 3318 and 3321. So anyway it appears that image sets that are composed of non-sequential image numbers throw the control point editor off... If you still need images to test, let me know. Thanks
Offline
If you can provide the image sets, I would be able to confirm this one solved with the new fisheye detection algorithm.
Offline