About blending issue in v2.5 compared to v2.0  

This forum is dedicated to Autopano Pro / Giga bug reports and features discussion.
Please read the posting rules before starting a topic!
User avatar
AlexandreJ
Kolor Team
 
Topic author
Posts: 5987
Likes: 7 posts
Liked in: 10 posts
Joined: Mon Nov 14, 2005 4:56 pm
Location: Francin, France
Info

About blending issue in v2.5 compared to v2.0

by AlexandreJ » Mon May 30, 2011 10:12 am

It seems that some people are having issue with a color harmonization in 2.5 engine that is not as good as in version 2.0.9.
I will try to make a resume of that issue here, give some hits and what is been done here.

Some quick facts :
- Autopano engine v2.0.9 was largely based on smartblend which does an anti-ghost seams detection + multiband blender. Smartblend does both algorithms on pairs of images ( 1+2, then result of (1+2) + 3, etc ).
- Autopano engine v2.5 is based on ideas and technics from smartblend, but revamped to address several issue smartblend had. First, the anti-ghost was doing it's job based on pairs decision. It's wrong because globally better solution can be found if you solve for ghost with all images at the same time.
Then the multiband can also be really faster if you do the blending with all image at the same time instead of doing the job 2 images at the same time.

So, globally, the new engine can do better job is many cases ( BTW : The Wembley image could not have been done without this new engine : http://www.kolor.com/blog-en/2011/05/18/autopano-giga-stitches-the-biggest-sports-gigapixel-photo-wembley-360/ ), but at the same it seems that some cases the harmonization in blue skies is less good than the old system.

First, some advice :
* Multiband default settings have changed between both version. If you setup by default -2 in the multiband level in engine v2.5, it's the value that emulates the most the behavior of the old engine.
* Color anchors have evolved also and this can change the global color harmonization of panorama too.

Here, we will work on this aspect of the blender to see what has really changed and how to get at least the same quality result if not better on these cases.
What we really need is to be able to get our hand on cases that illustrate that well. We already have some samples cases, but they are just a bit different, not really badder with the new version. If you have such cases, please, can you share them on autopano ftp : http://www.autopano.net/forum/t766-ftp-server

I will put this topic sticky until the full resolution of this issue. Be sure we are concerned about it and about the "auto" in autopano.

User avatar
taf
Member
 
Posts: 2684
Likes: 1 post
Liked in: 0 post
Joined: Tue Dec 20, 2005 12:13 am
Location: Paaaaaaris !
Info

by taf » Mon May 30, 2011 4:50 pm

Désolé, je répond en francais car c'est plus facile pour moi, vu la description...

OK, on met le paramètre du multiband à  -2 pour retrouver un rendu proche de l'ancien moteur, mais qu'est ce que l'on perd en faisant cela ?

je posterais ces prochains jours quelques panos sur lequel j'ai eu pas mal de pb sur les ciels...
A+
Look. There's a rhythmic ceremonial ritual coming up !

User avatar
AlexandreJ
Kolor Team
 
Topic author
Posts: 5987
Likes: 7 posts
Liked in: 10 posts
Joined: Mon Nov 14, 2005 4:56 pm
Location: Francin, France
Info

by AlexandreJ » Mon Jun 06, 2011 5:39 pm

On ne perd rien en fait, on evite des soucis dans certains cas que smartblend avait aussi. Les détails techniques sont vraiment assez complexes, je ne vais pas tout expliquer ici, mais sachez que la v2.6 sera mieux calibré que la version actuelle. Il risque d'en résulter un mélange plus lent, mais bon, si le résultat est meilleur.

User avatar
AlexandreJ
Kolor Team
 
Topic author
Posts: 5987
Likes: 7 posts
Liked in: 10 posts
Joined: Mon Nov 14, 2005 4:56 pm
Location: Francin, France
Info

by AlexandreJ » Mon Jun 06, 2011 5:39 pm

PS : je suis toujours très intéressé par tous cas qui marchent moins bien.

no avatar
lumelix
Member
 
Posts: 528
Likes: 0 post
Liked in: 0 post
Joined: Mon Oct 25, 2010 5:51 pm
Location: Switzerland
Info

by lumelix » Wed Jun 08, 2011 12:31 pm

Hello Alexandre

I send you a zip-file (aesch168_50p.zip) to your ftp server containing the 168 images (test subset) and pano-files of that case I describe in the bug-forum thread p83521 from 2011-05-23 14:15:38.
I have reduced the size of the images to 50% for fewer data
I hope that thereby you can better test the color correction system of APG 2,6 and APG 3.
Thank you for all helpful references ! I hope you can find a solution for this problem or give me some instruction to get better results.
Last edited by lumelix on Wed Jun 29, 2011 6:56 am, edited 1 time in total.
Regards
Martin

no avatar
GURL
Member
 
Posts: 2943
Likes: 0 post
Liked in: 0 post
Joined: Tue Dec 06, 2005 1:57 pm
Location: Grenoble
Info

by GURL » Tue Jun 28, 2011 4:56 pm

Here is an exemple where the same issues are are visible both in v2.5.2 Multiband and v2.5.2 Linear blenders.

This suggest that the problems could not depend on the used blender.
---

L'harmonisation dépend d'une part de la correction de couleur et des ancres d'autre part du blender. L'exemple ci-dessous semble suggerer que les différences entre les versions 2.0.8 et 2.5.2 ne provienne pas du blender puisqu'on retrouve les mêmes défauts (pas très visibles mais quand même visibles) avec Lineaire et avec Multiband


Last edited by GURL on Tue Jun 28, 2011 5:05 pm, edited 1 time in total.
Georges

User avatar
AlexandreJ
Kolor Team
 
Topic author
Posts: 5987
Likes: 7 posts
Liked in: 10 posts
Joined: Mon Nov 14, 2005 4:56 pm
Location: Francin, France
Info

by AlexandreJ » Wed Jun 29, 2011 8:09 am

One update : we did correct some part of the blender for the next coming 2.6 version. Some issues were coming from us and some are in fact due to the type of project.
We have done a tutorial to illustrate visual aspect of the new blender. What is cool with 2.5 engine is that it is really more powerfull but a little more complicate to adjust. That's why we did this tutorial. Coming soon with 2.6 version.

no avatar
gddxb
Member
 
Posts: 219
Likes: 0 post
Liked in: 0 post
Joined: Tue Apr 27, 2010 4:07 am
Info

by gddxb » Mon Aug 01, 2011 4:34 am

Hi Alexandre -

Not sure if related, but my findings as per thread may be of relevance. http://www.kolor.com/forum/t12721-blotchy-sky


Who is online

Users browsing this forum: No registered users and 3 guests