Image-stitching and virtual tour solutions My account Updates
It is currently Wed Aug 27, 2014 12:10 pm

All times are UTC + 1 hour




Post new topic Reply to topic  [ 18 posts ] 
Author Message
PostPosted: Wed Nov 03, 2010 3:03 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 13828
Location: Isleham, Cambridgeshire, UK.
Find the test image set here:
http://www.agnos.com/dati/download/samples/fce8-z1.zip

3 fisheye shots taken with Nikon P5100 with FC-E8 FE convertor, conversion factor 0.21.

With APG 2.5 beta 2
Sreenshots 1 and 2
After stitching, and rotating 180, the default pano orientation is such that a figure is bisected on the right edge. After rendering with Antighost and Spline 35 interpolator part of that figure remains as a ghost.
Screenshots 3 and 4
If I re-orientate the pano in the Panorama Editor so that the offending figure is not longer at the right edge then the ghost is removed.

With APG 2.0.9
Repeating the tests with matching pano orientations and rendering using APG 2.0.9, spline 26 interpolator and Smartblned blender.
Screenshots 5 and 6; and screenshots 7 and 8.

In neither instance is there a ghost.

Conclusion - Smartblend in APG 2.0.9 works better than the new Antighost blender in APG 2.5 beta 2.

(Can we have Smartblend back please?)

















_________________
Andrew Stephens
Nikon D40, Nikkor 10.5mm fisheye, Sigma 8mm f3.5 fisheye, Nikkor 18-55/50/35mm lenses, Nodal Ninja 5 Lite, Nodal Ninja 4 with R-D16, Agno's MrotatorTCS short.
Nikon P5100, CP5000, CP995, FC-E8, WC-E63,WC-E68, TC-E2, Kaidan Kiwi 995, Bophoto pano bracket, Agno's MrotatorA.
Merlin/Orion robotic pano head + Papywizard on Nokia 770/N800/N810 and Windows 8/XP/2K.


Last edited by mediavets on Thu Nov 04, 2010 1:42 pm, edited 1 time in total.

Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 4:43 pm 
Offline
Member
User avatar

Joined: Fri May 05, 2006 8:16 am
Posts: 1228
Location: Bulgaria
"Can we have Smartblend back please?"

:D :D :D Don't think that's possible, having in mind how much work went into rebuilding Smartblend from scratch! Still, I'd love to see an official response!

_________________
Some of my panoramas, posted in the Autopano Pro flickr group.


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:00 pm 
Offline
Member

Joined: Thu May 06, 2010 1:01 pm
Posts: 237
Location: Near Paris - France
A/ Can you really make a conclusion on one single pano ?

My understanding (and my experience with a "populated" pano) is that, from one run to another run of APG, the ghost removal may be different... So, maybe you were unlucky this time with APG 2.5 and lucky with APG 2.0.9...

B/ There is a trick to deterministically remove ghosts:

- save your .pano file !!!
- identify, among the images in your pano, the source image with the ghost you want to remove
- erase the ghost from this source image with gimp or photoshop (do not forget the alpha channel)
- Run APG with the same .pano file but with the modified image (if you changed the name of the image, edit the .pano file and change the name in it)

It works !

_________________
Nikon D5100 (formerly Nikon D60) - Sigma 10-20 - 24 shots Panoramas in 3 raws
Windows 32 bits - APG
website htt://jmh.trp.free.fr


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:04 pm 
Online
Administrator
User avatar

Joined: Mon Nov 14, 2005 4:56 pm
Posts: 5896
Location: Francin, France
I really hope that you are kidding, mediavets.
---
Nevertheless, this case is easy. The ghost is appearing on the edge. Just do a yaw + 50, render again and done.
---

case A : The antighost should be totally determistic in engine v2.5


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:31 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 13828
Location: Isleham, Cambridgeshire, UK.
AlexandreJ wrote:
I really hope that you are kidding, mediavets.

I'm just reporting my experience comparing the peformance of V2.5 Antighost and V2.0.9 with Smartblend with this image set.

It's plain that APG 2.0.9 with Smartblend performed better than APG 2.5 beta 2 Antighost in this example.

I think you'd have to agree?

Quote:
Nevertheless, this case is easy. The ghost is appearing on the edge. Just do a yaw + 50, render again and done.

Agreed - it's easy to fix - and that's what I did.

But my point is that I didn't have to do that with APG 2.0.9 and Smartblend; it didn't require fixing because it worked first time.

I don't think one should have to carefully inspect every rendering to check whether V2.5 Antighost has failed or not and then re-orientate the pano, if it has, and render again.

In a crowded scenes it's likely that one may have a figure on the edges of the pano whatever the yaw orientation - then what?

Does V2.5 Antighost have a problem with figures bisecting the edge of the stiched pano?

In other words what I am asking is why did V2.5 Antighost fail in this instance but V2.0.9 with Smartblend didn''t.

Are there things one needs to avoid to get Antighost to work reliably?
Quote:
case A : The antighost should be totally determistic in engine v2.5

I don't understand what you mean by this sentence - the meaning seems to have been 'lost in translation'.

_________________
Andrew Stephens
Nikon D40, Nikkor 10.5mm fisheye, Sigma 8mm f3.5 fisheye, Nikkor 18-55/50/35mm lenses, Nodal Ninja 5 Lite, Nodal Ninja 4 with R-D16, Agno's MrotatorTCS short.
Nikon P5100, CP5000, CP995, FC-E8, WC-E63,WC-E68, TC-E2, Kaidan Kiwi 995, Bophoto pano bracket, Agno's MrotatorA.
Merlin/Orion robotic pano head + Papywizard on Nokia 770/N800/N810 and Windows 8/XP/2K.


Last edited by mediavets on Wed Nov 03, 2010 5:32 pm, edited 1 time in total.

Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:37 pm 
Offline
Member
User avatar

Joined: Fri May 05, 2006 8:16 am
Posts: 1228
Location: Bulgaria
I think Alexandre means in v2.5 you should be able to predict (and control?) the outcome of a blend by looking at the panorama preview. And it should be repeatable, no random factor - so if that person was "ghosted-out" in this panorama detection, it should be ghosted-out if you re-render (or re-detect) the same image set.

I'm still wondering if it will be possible to produce a near-real-time preview in Panorama Editor with anti-ghosting or the same settings as the final render, only reduced resolution, say 1200x800.

_________________
Some of my panoramas, posted in the Autopano Pro flickr group.


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:38 pm 
Online
Administrator
User avatar

Joined: Mon Nov 14, 2005 4:56 pm
Posts: 5896
Location: Francin, France
mediavets wrote:
I'm just reporting my experience comparing the peformance of V2.5 Antighost and V2.0.9 with Smartblend with this image set.

Sure, in this example the old smartblend works better. But I think I found the issue in the new engine in fact.
We forget to put back the code that handles ghost over 360° edges. This code were in the 2.0.9 engine and probably forgot in the new engine.

Issue 405 opened


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:42 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 13828
Location: Isleham, Cambridgeshire, UK.
AlexandreJ wrote:
mediavets wrote:
I'm just reporting my experience comparing the peformance of V2.5 Antighost and V2.0.9 with Smartblend with this image set.

Sure, in this example the old smartblend works better. But I think I found the issue in the new engine in fact.
We forget to put back the code that handles ghost over 360° edges. This code were in the 2.0.9 engine and probably forgot in the new engine.

Issue 405 opened

OK - so just maybe it was worth my reporting my experiences? ;)

_________________
Andrew Stephens
Nikon D40, Nikkor 10.5mm fisheye, Sigma 8mm f3.5 fisheye, Nikkor 18-55/50/35mm lenses, Nodal Ninja 5 Lite, Nodal Ninja 4 with R-D16, Agno's MrotatorTCS short.
Nikon P5100, CP5000, CP995, FC-E8, WC-E63,WC-E68, TC-E2, Kaidan Kiwi 995, Bophoto pano bracket, Agno's MrotatorA.
Merlin/Orion robotic pano head + Papywizard on Nokia 770/N800/N810 and Windows 8/XP/2K.


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:43 pm 
Offline
Member
User avatar

Joined: Fri Sep 22, 2006 7:58 am
Posts: 277
Location: Chambéry
If you see my topic : give us back the good old smartblend.........


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 5:57 pm 
Online
Administrator
User avatar

Joined: Mon Nov 14, 2005 4:56 pm
Posts: 5896
Location: Francin, France
mediavets wrote:
OK - so just maybe it was worth my reporting my experiences? ;)

Do report, always. It my job to say, it is relevant or not. This one was because of the 360° edge.

To mountainb. I saw your thread and it was transfered for deeper analysis to a developer.


Top
 Profile  
 
 Post subject:
PostPosted: Wed Nov 03, 2010 8:37 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 13828
Location: Isleham, Cambridgeshire, UK.
I have another example of an apparent failure of V2.5 Antighost this time not at the edge of a pano.

The image set can be found here:
http://www.agnos.com/dati/download/samples/frarihi.zip

3 shots on D100 with Sigma 8mm fisheye, shot with 60 degree roll.

APG V2.5 beta 2 with spline 36 and Antighost

Screenshots 1 and 2.

APG 2.0.9 with spline 36 interpolator and Smartblend

Screenshots 3 and 4.

......

Colour correction is also horrible with the rendered pano from APG 2.5 b2.

Screenshot 5.

This appears to be an instance of this issue?:
http://www.autopano.net/forum/t10408-problem-of-dirty-zone-after-rendering











_________________
Andrew Stephens
Nikon D40, Nikkor 10.5mm fisheye, Sigma 8mm f3.5 fisheye, Nikkor 18-55/50/35mm lenses, Nodal Ninja 5 Lite, Nodal Ninja 4 with R-D16, Agno's MrotatorTCS short.
Nikon P5100, CP5000, CP995, FC-E8, WC-E63,WC-E68, TC-E2, Kaidan Kiwi 995, Bophoto pano bracket, Agno's MrotatorA.
Merlin/Orion robotic pano head + Papywizard on Nokia 770/N800/N810 and Windows 8/XP/2K.


Last edited by mediavets on Wed Nov 03, 2010 8:46 pm, edited 1 time in total.

Top
 Profile  
 
 Post subject:
PostPosted: Thu Nov 04, 2010 8:59 am 
Online
Administrator
User avatar

Joined: Mon Nov 14, 2005 4:56 pm
Posts: 5896
Location: Francin, France
Interesting case. But this time, it's not that relevant :)
You did a panorama with a really wrong color correction. Or this has an influence on the antighost algorithm, because that algorithm is based on a rule that try to find path that will have the lowest visual difference. Or if these differences are modified by a too strong color cast ( coming from color correction in this case ), it won't find the right path.

I redid the same panorama without color correction and with a good color correction, and both leads to the right rendering.

Let's keep that in mind : badly color corrected panorama can change the result of the antighost algorithm.


BTW : Some people are seems some issues with the new antighost. How many of you are using it in complicated cases where we can clearly see that it's better ?




Top
 Profile  
 
 Post subject:
PostPosted: Thu Nov 04, 2010 10:15 am 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 13828
Location: Isleham, Cambridgeshire, UK.
AlexandreJ wrote:
Interesting case. But this time, it's not that relevant :)
You did a panorama with a really wrong color correction. Or this has an influence on the antighost algorithm, because that algorithm is based on a rule that try to find path that will have the lowest visual difference. Or if these differences are modified by a too strong color cast ( coming from color correction in this case ), it won't find the right path.

I redid the same panorama without color correction and with a good color correction, and both leads to the right rendering.

Let's keep that in mind : badly color corrected panorama can change the result of the antighost algorithm.

OK - so colour correction and antighost interact and are in some way mutually dependent? So that if colour correction is not set/working properly then antighost will fail. Is that a corrcet imterpreration of you comments?

Well....I used the default settings - see screenshot below.

I assumed that the default colour correction settings would be most appropriate for most panos? Perhaps not?

How did my colour settings become 'really wrong' when I was using the defaults? Are the defaults set incorrectly? Have I inadvertently changed the defaults without having been aware of doing so?

What colour correction should I have set to get this scene to stitch properly as you have done?

Does one have to change the sdetting for colour correction depending on some pre-stitch assessment of the image set? If so, how? What are the rules/metrics one should employ to ensure reliable results?

How do all those possible colour correction settings relate to None/LDR/HDR? That system I more or less understood, or rather I became familiar with how it behaved. This new system I find quite incomprehensible and quite unpredicatble to the point of being unsable and many times the defaults produce completely bizarre and unacceptable results, which are nowhere near as good as I got from APG 2.0.9 without any 'monekeying about'.

I have had many instances where colour correction didn't seem to work at all and I just ended up with a brown/grey mess; screenshots posted by others suggest that I am not alone in having this experience.

And if the colour correction gets screwed up it seems that any attempt to correct things merely makes things worse, whereas I could usually sort it out with the old colour anchor system. And I don't think I'm alone in that.

.......

What are the 'tiricks' one must employ to get colour correction to work properly?

My impression is that colour correction is 'broken' in beta2 - if you think it is not, then please tell me how to get it to work correctly and reliably.



_________________
Andrew Stephens
Nikon D40, Nikkor 10.5mm fisheye, Sigma 8mm f3.5 fisheye, Nikkor 18-55/50/35mm lenses, Nodal Ninja 5 Lite, Nodal Ninja 4 with R-D16, Agno's MrotatorTCS short.
Nikon P5100, CP5000, CP995, FC-E8, WC-E63,WC-E68, TC-E2, Kaidan Kiwi 995, Bophoto pano bracket, Agno's MrotatorA.
Merlin/Orion robotic pano head + Papywizard on Nokia 770/N800/N810 and Windows 8/XP/2K.


Last edited by mediavets on Thu Nov 04, 2010 10:21 am, edited 1 time in total.

Top
 Profile  
 
 Post subject:
PostPosted: Thu Nov 04, 2010 11:10 am 
Online
Administrator
User avatar

Joined: Mon Nov 14, 2005 4:56 pm
Posts: 5896
Location: Francin, France
mediavets wrote:
OK - so colour correction and antighost interact and are in some way mutually dependent? So that if colour correction is not set/working properly then antighost will fail. Is that a corrcet imterpreration of you comments?

That was always the case even with 2.0 series, but don't misunderstand me. There is a relation because of the following fact :
Imagine a pure light blue sky in one image and a dark pure blue sky in a second image. These images are overlapping. For the antighost system, it will find that there is a ghost between both images because of the difference of color.

It's exactly the same with this case. The wrong color correction probably changed the decision taken by anti-ghost ( because the ground on one image what light grey and on the second dark grey. Perhaps one moving people was having the same color as the ground of the second image, etc ).

It's about pixels, not about enabled / disabled color correction. In the following case, I just disabled color correction because it was not needed. And it worked directly. In a really badly exposed panorama, you will need to make color correction to help anti-ghost to work better.


mediavets wrote:
I assumed that the default colour correction settings would be most appropriate for most panos? Perhaps not?
How did my colour settings become 'really wrong' when I was using the defaults? Are the defaults set incorrectly? Have I inadvertently changed the defaults without having been aware of doing so?

I still have to check that (issue 390, btw)

mediavets wrote:
What colour correction should I have set to get this scene to stitch properly as you have done?

No color correction.

mediavets wrote:
Does one have to change the sdetting for colour correction depending on some pre-stitch assessment of the image set? If so, how? What are the rules/metrics one should employ to ensure reliable results?

No. Most of the time, you don't need to cope with that. We are in a theorical case here : how to make anti-ghost work when color are obviously wrong ?
So just check color correction ( by disabling or changing anchor type ) and everything fine.

mediavets wrote:
How do all those possible colour correction settings relate to None/LDR/HDR? That system I more or less understood, or rather I became familiar with how it behaved. This new system I find quite incomprehensible and quite unpredicatble to the point of being unsable and many times the defaults produce completely bizarre and unacceptable results, which are nowhere near as good as I got from APG 2.0.9 without any 'monekeying about'.

I have had many instances where colour correction didn't seem to work at all and I just ended up with a brown/grey mess; screenshots posted by others suggest that I am not alone in having this experience.

What are the 'tiricks' one must employ to get colour correction to work properly?

I solved a bug for RC that is 358 ( broken exposure and gamma correction ), so you cannot use this type of anchors.
There is another bug where the imported setting from beta 1 is doing issue. So put 'gamma' as default value, restart and that will also solves a lot of issue already.


Top
 Profile  
 
 Post subject:
PostPosted: Thu Nov 04, 2010 12:46 pm 
Offline
Member
User avatar

Joined: Fri May 05, 2006 8:16 am
Posts: 1228
Location: Bulgaria
I thing a big part of the problem is the color correction bug in beta2. Andrew is right that there's a problem, however the whole "new smartblend sucks / old rocks / v2.5 is stupid"-thing is based on testing a beta version! Many of the heated comments in this and other threads seem to disregard this simple fact. Which makes me nervous :)

_________________
Some of my panoramas, posted in the Autopano Pro flickr group.


Top
 Profile  
 
 Post subject:
PostPosted: Thu Nov 04, 2010 1:39 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 13828
Location: Isleham, Cambridgeshire, UK.
'[bo wrote:
']I thing a big part of the problem is the color correction bug in beta2. Andrew is right that there's a problem, however the whole "new smartblend sucks / old rocks / v2.5 is stupid"-thing is based on testing a beta version! Many of the heated comments in this and other threads seem to disregard this simple fact. Which makes me nervous :)

On the whole I agree.

It was very silly of me to ask for Smartblend back - what I meant of course was that I wished the new Blender in V2.5 would perform as reliably and consistently as Smartblend did in V2.0.9.

That said it's plain it doesn't yet for whatever reason. Perhaps because of the interaction with colour correction and what appear to be unresolved issues with the new colour correction system. So for the user - as opposed to the developer - what appear to be antighost failures may in fact be attributable to colour correction problems. When you add the new exposure blending and stacking features into the mix it can become very difficult (for the user) to have any clear idea of what may be causing the unexpected and undesirable behaviours one observes; systematic testing becomes very complex.

If there was a concise list of unresolved issues (and their implications) published with the release of beta 2 then I've overlooked it. I think there was a list of known bugs that had been fixed - perhaps it would have been qually useful to have a list of known bugs that had not yet been fixed? That way we'd know what to expect NOT to work yet and it would save us perhaps from repeatedly reporting already known bugs.

The recent instances of code inadvertently being left out of beta builds of both APP/APG and Panotour, and this only being discovered as a result of users reporting 'bugs', suggests that there is also some scope for improvement in Kolor's software testing, build control and QA processes now that the software has become so much more complex and feature-rich. I'm sure they have enough to do without investigating and fixing 'bugs' that are already fixed.

I suppose I expected beta 2 to perform better than it does, but with so many new features in V2.5 I guess it's understandable that unexpected and unpredictable behaviours and interactions will emerge in beta testing as more and more users try the software, as we all have different ways of working with the software. I think open alpha and beta testing is a good thing - the more users who try the software the more likely it is that bugs will be uncovered sooner rather than later. Such user involvement can also improve useability (and has often done so in the past with Kolor products).

I'm sure APP, APG and Panotour will get better and better and I look forward to the the next releases.

_________________
Andrew Stephens
Nikon D40, Nikkor 10.5mm fisheye, Sigma 8mm f3.5 fisheye, Nikkor 18-55/50/35mm lenses, Nodal Ninja 5 Lite, Nodal Ninja 4 with R-D16, Agno's MrotatorTCS short.
Nikon P5100, CP5000, CP995, FC-E8, WC-E63,WC-E68, TC-E2, Kaidan Kiwi 995, Bophoto pano bracket, Agno's MrotatorA.
Merlin/Orion robotic pano head + Papywizard on Nokia 770/N800/N810 and Windows 8/XP/2K.


Last edited by mediavets on Thu Nov 04, 2010 2:37 pm, edited 1 time in total.

Top
 Profile  
 
 Post subject:
PostPosted: Thu Nov 04, 2010 4:15 pm 
Offline
Member

Joined: Thu May 03, 2007 9:43 pm
Posts: 198
Location: the Netherlands
I also agree with bo, I'm having the same issues as mountaintb regarding the dirty areas.

For the better part APG v2.5b2 is a great improvement from v2.09, detection is better, optimization with multiple viewpoints is very effective. Also the anti-ghosts multiblend feature is excellent for removing ghosts in parts with great detail. However, blending large areas without detail (a blue sky) is where it falls short. Smartblend is exelent for blending large areas without much detail, but falls short when blending fine detail and continuous lines.

This pano I was working on this afternoon for example, APG v2.09 blends the sky perfectly but there is a small misalignment in the metal wires of the bridge (not visible on this scale, check the full resolution pano). In the APG v2.5b2 the problems are obvious, the images are not blended smoothly and check out the grey area on the left. I've tried all possible multiblend level settings, with or without color correction, the result is never as good as the v2.09 version.

krpano of APG v2.09... http://ronaldtichelaar.nl/forumposts/autopano/apg209/

krpano of APG v2.5b2... http://ronaldtichelaar.nl/forumposts/autopano/apg25b2/

Also in the interactive versions it's visible where the ends join to become a sphere. It's the first time I've notices this in APG v2.5b2, I'm also seeing it in the v2.09 version but much less noticeable. (images are only converted 16bit > 8bit, Adobe RGB > sRGB, save as jpg, no other processing)







_________________
Ronald

Fujifilm S5 Pro, Nikon D300, NodalNinja 3mk2, Sigma 10mm f/2.8 fisheye HSM,
Sigma 30mm f/1.4 EX HSM, Sigma 24-70 f/2.8 EX HSM, Sigma 70-200mm f/2.8 EX HSM


Top
 Profile  
 
 Post subject:
PostPosted: Mon Nov 08, 2010 5:12 pm 
Online
Administrator
User avatar

Joined: Mon Nov 14, 2005 4:56 pm
Posts: 5896
Location: Francin, France
There are two issues on the same thread here. Let's separate them :
- 360° ghost issue => this is solved for RC version ( that's the initial topic for this thread )
- the blending quality => I've opened the issue 411. let's continue this discussion on the right thread :
http://www.autopano.net/forum/t10432-apg-v2.5b2-multiblend-antig-host-renders-different-at-90a-pitch


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 18 posts ] 

All times are UTC + 1 hour


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Group