4

I recently tried to make a "360 photo" out of a series of images. I put my tripod in the center of the room and took almost 300 photos, with plenty of overlap (35mm lens on a Nikon DX body).

Then I went and loaded them in Hugin. It started fine, but then it wanted to match every image to every other image. So i decided to reduce the resolution of the photos (from 6 to 1 megapixels), and only process half of the photos. So I discarded all the odd number filenames. This made the process much quicker but when it got to the panorama preview... it was a mess. A lot of mismatched points, images upside down, points not detected. I messed around for a few hours, trying different combinations of settings, and it was no better.

So after adding a few more points manually the panorama started to take shape, but it was still a mess.

So i just downloaded the trial version of Autopano. Installed it and in less than 2 minutes it processed the full image set (300 images) and made a perfect panorama on the first try.

What's going on here? Did I use a different profile, setting, or is it just the algorithm on Autopano being superior to Hugin's?

(FWIW: I am familiar with feature detection and matching. I am the person who made this video, where I match features in real time against 20 thousand images only using a Core 2 Duo)

  • Not sure about the technical details of Autopano vs Hugin, but I'd easily go with superior proprietary(?) algorithms, as is usually the case with expensive vs FOSS software (when the former is better). – JoséNunoFerreira Oct 7 '16 at 17:26
  • The algorithm plays an important role here for sure. Additionally Hugin is a free tool. I'd say one can't expect so much from a free application but hopefully there are some guys out there who contribute to these opensource applications and make it better through time. – SERAJ Jan 19 '17 at 20:38
  • Is it a 360 degree panorama (typically from one row of pictures horizontally) or a spherical photo (aka virtual tour)? Btw. with a 35mm lens on an APS-C sensor those 300 pics are some redundancy! – user681768917 May 25 '17 at 14:21
  • @user681768917 it's a spherical photo – hjf May 25 '17 at 14:23
  • Don't know the answer 'why' but try even less pictures. In my experience Hugin get's confused if there are too many redundant pictures - as if it didn't know what to do with them. Not sure about spherical photos, but you can try to find and connect another program for finding control points to Hugin. – user681768917 May 25 '17 at 14:27
1

I could be trite and say: better QA (debugging) process and algorithmic improvements... but I can be much more specific:

Two reasons (and I've seen the same effect BTW, having recently cleaned up a multi-row 360 pano shot with 70-200 Nikon and FX sensor.)

1) Hugin definitely attempts to match every photo to every other photo. That's not necessarily a problem except that it can require quite a lot of memory and scratch disk space. One bug/limitation of Hugin is that it doesn't monitor free space well and doesn't overflow temp file usage to other partitions. YES, reducing resolution for the matching can help. But then you've got to go through and clean up the matches.

2) Hugin definitely gets confused at times, not only about significant overlaps, but also about the specific order in which they are presented. For example, suppose you have a row sequence like this, with significant overlap: 1 2 3 4 5

I have found many cases where even though everything appears to be fine, photo #3 will be improperly integrated into the pano (and/or the whole thing may crash etc)... yet if I simply resequence so #3 is "on top" then all is well: 1 2 5 3 4 (ie the order in the list becomes P1,P2, P4,P5,P3 or even 1 2 4 3 5 (ie just move P3 over one: P1, P2, P4, P3, P5)

Obviously, if these kinds of bugs had been fully diagnosed, Hugin wouldn't have these issues.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.