Page 1 of 1

Blending - reproducing Smartblend results in v2.5

PostPosted: Sun Feb 27, 2011 6:42 am
by rjllane
I eagerly installed v2.5 having been very impressed with v2.0.9. I like the new interface, and haze reduction plugin. However, I have not been able to achieve blending results as good as those that I got with the default Smartblend option in v2.0.9. As far as I can see, Smartblend is no longer an option.

Is this correct?

If so, can anyone suggest how I can reproduce the results using the options and settings available in v2.5?

My principle issue is that the blending in v2.5 appears to be very harsh and abrupt using v2.5 default settings. It almost looks like the adjacent images are simply masked and put together without any blending at all. What has been the experiences of other users?

PostPosted: Sun Feb 27, 2011 3:22 pm
by klausesser
rjllane wrote:I eagerly installed v2.5 having been very impressed with v2.0.9. I like the new interface, and haze reduction plugin. However, I have not been able to achieve blending results as good as those that I got with the default Smartblend option in v2.0.9. As far as I can see, Smartblend is no longer an option.

Is this correct?

If so, can anyone suggest how I can reproduce the results using the options and settings available in v2.5?

My principle issue is that the blending in v2.5 appears to be very harsh and abrupt using v2.5 default settings. It almost looks like the adjacent images are simply masked and put together without any blending at all. What has been the experiences of other users?

Could you post a screenshot of your render-settings?

best, Klaus

PostPosted: Sun Feb 27, 2011 8:05 pm
by Bmachine
Yes, I am getting the same results.

I have posted a thread in the Autopano bug & discussion forum because that is where these issues should be reported.

Bo

PostPosted: Mon Feb 28, 2011 12:49 pm
by rjllane
Hello Klaus and others.

Following are typical examples comparing Autopano Pro v2.0.09 (given second) with Autopano Giga v2.5.0 (given first).

The procedure was the same in both examples, utilising the out-of-the-box default settings; identify 4 images, detect, render. The 4 images occupy a simple 2 rows by 2 columns arrangement. The snapshots of the rendered output are at 1:1 from the centre of the panorama where all 4 images join.

The Autopano Giga v2.5.0 output has flat colouring, and you can see blurred/fuzzy regions all around the centre of the panorama where all 4 images join. Whatever has changed from v2.0.09 to v2.0.5, the default settings would suggest that I am better off staying with v2.0.09. What changed? Are there settings in v2.0.5 that reproduce the v2.0.09 results? As indicated, I like the new interface and the haze filtering plugin, and would like to use v2.5.0 to access these features, but not if the output is consistently as poor as this compared to the previous version.

PostPosted: Mon Feb 28, 2011 1:36 pm
by AlexandreJ
rjllane wrote:The Autopano Giga v2.5.0 output has flat colouring, and you can see blurred/fuzzy regions all around the centre of the panorama where all 4 images join. Whatever has changed from v2.0.09 to v2.0.5, the default settings would suggest that I am better off staying with v2.0.09. What changed? Are there settings in v2.0.5 that reproduce the v2.0.09 results? As indicated, I like the new interface and the haze filtering plugin, and would like to use v2.5.0 to access these features, but not if the output is consistently as poor as this compared to the previous version.

Interesting example because it illustrate something totally unrelated with rendering : the color correction between both version have changed.
With this case, it could be nice to post the same rendering zone without color correction in both version of the software ( 2.0.9 and 2.5 ).

BTW : A quick sum up of what have changed in the engine and how it could influence the result :
- New multiband blender => change the way blending is done. Should be better in 2.5.
- New anti-ghost decision => the decision made by 2.0 and 2.5 can be diffferent, but globally, 2.5 has been measured to give really better results in high speed moving scene ( like crowd for example )
- Color correction : the system is quite the same but optimization is better and default anchor changed. It can change illumination of the scene if we compare both version.
- Better optimization => globally, a stitching with 2.5 has a better geometrical matching but it could depend of the type of stitch ( multiple viewpoint is a bit different ).

PostPosted: Mon Feb 28, 2011 10:18 pm
by rjllane
Hello AlexandreJ and others.

Can you please clarify what you mean by "without color correction"?

"- New multiband blender => change the way blending is done. Should be better in 2.5."
Perhaps the multiband blender is better. The example I gave here was not exactly what I have seen in other examples (which I will hunt for when I get a chance). Why not have the previous Smartblend process as an option?

"- New anti-ghost decision => the decision made by 2.0 and 2.5 can be diffferent, but globally, 2.5 has been measured to give really better results in high speed moving scene ( like crowd for example )"
I have not put this to the test yet.

"- Color correction : the system is quite the same but optimization is better and default anchor changed. It can change illumination of the scene if we compare both version."
In this instance, v2.5 is worse, not better. I have yet to see where it was better. What have other users reading this forum experienced?

"- Better optimization => globally, a stitching with 2.5 has a better geometrical matching but it could depend of the type of stitch ( multiple viewpoint is a bit different )."
I have used the the default settings in this example, and I would attribute the blurring around the join in v2.5 to inferior geometrical matching.

In summary, I am yet to see an instance where v2.5 is even equal to v2.0.09 in quality. Can we look forward to a v2.5.01 release where the previous methods for blending, color correction, and geometrical matching are restored as options, even if they are not the defaults?

PostPosted: Tue Mar 01, 2011 2:04 am
by kzaugg
I am having a similar experience.

This picture works fine in 2.0.6, but not in 2.5

Image

PostPosted: Tue Mar 01, 2011 11:56 am
by klausesser
rjllane wrote:What have other users reading this forum experienced?

It´s definitely better. I realized that the settings are somewhat different - and due to the lack of ANY documentation that might rise problems.
That´s related to nearly all functions in APG.

best, Klaus

PostPosted: Tue Mar 01, 2011 1:21 pm
by AlexandreJ
Documentation translation is quite finish. We'll put it online next week probably.

PostPosted: Wed Mar 02, 2011 11:40 am
by [bo]
rjllane wrote:Can we look forward to a v2.5.01 release where the previous methods for blending, color correction, and geometrical matching are restored as options, even if they are not the defaults?

No.

:D

"The previous methods" were modified and/or rewritten precisely to improve them the ways already described. It'd be silly (if not impossible) to re-introduce old code. I, for one, would not want such a step backwards.

"The previous Smartblend process" was really problematic and slow, even with small panos. I have yet to see a pano of my own material that blends worse in 2.5, compared to 2.0. In your example, v2.5 is actually different, not worse. You main "problem" is with global illumination and the way color correction in 2.5 works, not the actual blending.

What color anchors have you set?

PostPosted: Wed Mar 02, 2011 3:34 pm
by Bmachine
I have been having similar issues. I discussed them in the bugs forum.

http://www.autopano.net/forum/viewtopic.php?pid=78507#p78507

What finally solved the problem for me was to trash the preference files. Here is how Alexandre describes it:

AlexandreJ wrote:On windows, you can delete Kolor registry : HKEY_CURRENT_USER/Software/Kolor/Autopano (Pro or Giga)2.5/
On mac, In the ~/Library/Preferences/, remove all fichiers starting with com.kolor.

Bo