Image-stitching and virtual tour solutions My account Updates
It is currently Wed Oct 22, 2014 9:20 am

All times are UTC + 1 hour




Post new topic Reply to topic  [ 17 posts ] 
Author Message
PostPosted: Tue Aug 05, 2014 4:41 pm 
Offline
New member

Joined: Tue Aug 05, 2014 4:19 pm
Posts: 6
We have discovered a bug with printing out HDR files (probably other types too)
from a stitching of 3 fisheye image sets.
There is a black line (bad interpolation) on the right hand edge of all
panormaic renders we are doing with the lates Autopano Giga 64bit for Linux.

See example edge image viewed with Luminance-HDR,

http://www.pasteall.org/pic/show.php?id=74962

Also really visible in 360deg view within our 3D software (blender),

http://www.pasteall.org/pic/show.php?id=74963

I have archived my stitching example so you can reproduce it,
please download this file,

http://brunnur.vedur.is/pub/hroi/Autopa ... bug.tar.gz

It has the autopano workspace file and all relevant fisheye images.
The dark line bug becomes very apparent when using -6 to -10 in blending mode: multiband,
and is always present when using blending mode: linear.

During default blending mode , "-2" the effect is hardly visible,
but we require more linear combination of the bracketed exposures
for our 3D VFX work.

We feel this bug is a serious bug with the current Autopano Giga for 64bit Linux,
and it may also affect other version we have not tested it.
It makes our 360deg stitching scenes unusuable and we cannot service
our customers without some hacking and editing of the resulting HDR image.

Please, if you can assist us with giving us a link to older autopano versions that do not
have this bug, that would be very very useful for us right now - or develop a fix
ASAP.

Btw, we are very impressed and happy with autopano giga. It is a very very functional,
full of features for our work in VFX.

Many thanks in advance,
Hrob.
-------------------------------
Hrobjartur Thorsteinsson
SanD Visual
www.sandvisual.com


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 5:33 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 14038
Location: Isleham, Cambridgeshire, UK.
thorsteinssonh wrote:
Please, if you can assist us with giving us a link to older autopano versions that do not
have this bug, that would be very very useful for us right now - or develop a fix
ASAP.


You can always find earlier versions of Kolor software here:

http://www.kolor.com/download-previous-version.html


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 6:26 pm 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
thorsteinssonh wrote:
We have discovered a bug with printing out HDR files (probably other types too)
from a stitching of 3 fisheye image sets.
There is a black line (bad interpolation) on the right hand edge of all
panormaic renders we are doing with the lates Autopano Giga 64bit for Linux.



Hi!

Let´s sort out some things.

1) i guess you mean "render" instead of "print out"!?

2) How did you shoot the "3 fisheye image sets" Can you go into details?

3) A seam usually occurs by having not 2:1 render-size - so what´s your render-size precisely?

4) Which way do you feed it to a 3D software?

I definitely do not think that this is a "bug" - i think you made a mistale somewhere. Let´s try to find out where.
I´m using equirectangular image files, made with APG, quite often for Cinema4D and Maya. I never had this issue which
you describe.

I´ll first hand check your images.

Klaus

PS: downloading elder APG versions would most unlikely be of any help.


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 6:33 pm 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
Having seen your site i guess you´re talking of HDR-probes for IBL?

Dis you shoot strictly spherical? Why only 3 shots for a 360x180° sphere?

Which lens did you use and which camera and which head?

Klaus


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 6:52 pm 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
the very firste test seem to indicate an issue with the lens-settings.

Attachment:
Bildschirmfoto 2014-08-05 um 19.50.35.png
Bildschirmfoto 2014-08-05 um 19.50.35.png [ 2.53 MiB | Viewed 636 times ]


Klaus

PS:
why do you have 9 steps of bracketing per position from which 4 are identical?
Please tell us about your shooting-strategy!

PS.:2

Ii definitely need to know more about the way you shoot. I tried your shots in Photomatix - weird results. Never saw those.

PS.:3

This is really kind of strange:

Attachment:
Bildschirmfoto 2014-08-05 um 20.28.03.png
Bildschirmfoto 2014-08-05 um 20.28.03.png [ 172.99 KiB | Viewed 633 times ]


Attachment:
Bildschirmfoto 2014-08-05 um 20.34.37.png
Bildschirmfoto 2014-08-05 um 20.34.37.png [ 268.14 KiB | Viewed 633 times ]


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 10:00 pm 
Offline
New member

Joined: Tue Aug 05, 2014 4:19 pm
Posts: 6
Dear Klaus, thank you kindly for investigating our issue. In reply to your questions,
You are correct, our issue occurs during rendering, and the purpose is IBL. Again you are right that we have uncommon lens setup. Our lens is the Sunex Super Fisheye, See, http://www.superfisheye.com

It is extremely wide so only 3 angles necessary. We did not observe the rendering probleme last year before we went through upgrades last week.

You noticed something i will have to check out, perhaps some of the brackets are repetitions and we did not realise this. I will check this out later at my workstation and get back to you.


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 10:20 pm 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
thorsteinssonh wrote:
Dear Klaus, thank you kindly for investigating our issue. In reply to your questions,
You are correct, our issue occurs during rendering, and the purpose is IBL. Again you are right that we have uncommon lens setup. Our lens is the Sunex Super Fisheye, See, http://www.superfisheye.com

It is extremely wide so only 3 angles necessary. We did not observe the rendering probleme last year before we went through upgrades last week.

You noticed something i will have to check out, perhaps some of the brackets are repetitions and we did not realise this. I will check this out later at my workstation and get back to you.



Hi Thorsten!

Afaik tge Sunnex has a somewhat "special" kind of projection - i´m not sure, but will go after it.

Basically it´s not a too good idea using such extremely short fisheyes. As you can see on your images they only fill a very small part of the image: a rather small circle.

The problem: you loose MUCH of the resolution the sensor provides. For that reason the resulting image-quality isn´t as good as it could be. Using it "just" for IBL-probes that doesn´t count too much: usually they need to be a bit blurry anyway.

But the advantage you can take from higher resolution is a better/sharper mirror-map and the ability to use parts of the sphere as backplates. That´s why i use a D800 and a 16mm fisheye.

Using a D800/16mm fisheye or a 5D2/15mm fisheye provides 112Megapixels and the D800 provides 180Megapixels.

In both cases it takes 7 shots - 6@-12° and one Zenith @90. They´re done in about 45 secondes using a NodalNinja.

Related to your camera i suggest a 10,5mm Nikon fisheye for the Canon (adapter round 70.-€). This also takes 7 shots - but provides much better resolution, is a well known and supported lens - and produces less CAs @f8.

best and good luck, Klaus


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 10:26 pm 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 14038
Location: Isleham, Cambridgeshire, UK.
klausesser wrote:
Related to your camera i suggest a 10,5mm Nikon fisheye for the Canon (adapter round 70.-€). This also takes 7 shots - but provides much better resolution, is a well known and supported lens - and produces less CAs @f8.

best and good luck, Klaus


That odd combination doesn't offer aperture control as I recall?

I think nowadays people would prefer the Sigma 10mm fisheye or the Tokina 10-17mm zoom fisheye with Canon mount?


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 11:46 pm 
Offline
New member

Joined: Tue Aug 05, 2014 4:19 pm
Posts: 6
Thanks guys for the lens suggestions. I will definately investigate these options.
Our Sunex choice has been sufficient and fast solution on location, and simple mirror ball images are fine too.
I think we will eventually need higher res evironment images as you suggest in some situations.

Do you know if there is a software bug tracker online for Autopano?
Blog type bug reporting is a bit unusual to me, and doesn't focus software-wise on the issue we have discovered.
I take it most people commenting here are not working for Kolor on developing the Autopano interface and functionality?


Top
 Profile  
 
PostPosted: Tue Aug 05, 2014 11:47 pm 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
mediavets wrote:
klausesser wrote:
Related to your camera i suggest a 10,5mm Nikon fisheye for the Canon (adapter round 70.-€). This also takes 7 shots - but provides much better resolution, is a well known and supported lens - and produces less CAs @f8.

best and good luck, Klaus


That odd combination doesn't offer aperture control as I recall?


"odd combination"?? I used it myself successfully for quite a time.

For what would you need an aperture control on a fisheye?

mediavets wrote:
I think nowadays people would prefer the Sigma 10mm fisheye or the Tokina 10-17mm zoom fisheye with Canon mount


When i said Nikon 10,5mm that 1) was an example regarding the focal-length and resolution and 2nd) because i know this lens by working with it on a 20D some years ago. Excellent fisheye!

You´re experienced yourself using the Sigma 10mm FE or the Tokina 10-17mm zoom fisheye for panoramas?

Klaus


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 12:19 am 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
thorsteinssonh wrote:
Do you know if there is a software bug tracker online for Autopano?
I take it most people commenting here are not working for Kolor on developing the Autopano interface and functionality?


I´m quite sure the issue isn´t a "bug". There are fisheye-projection models which are commonly used because most of the lenses available use the one or two certain projection-methods.

The Sunex is an extremely "exotic" lens - which needs it´s own dedicated application for undistortion and has a certain and uncommon projection-model.

This all isn´t of any problem as long as you use it for usual photography - but might cause trouble when it comes to stitching.

Whether Kolor has it in it´s database or not i don´t know. But i know that commercial/professional photographers - like myself - usually avoid using exotical gear
but rely on well proven devices.

It might figure out that the problem is this very special type of lens for which no lens-model
exists so far.

But i don´t know. What i know: its very unlikely a bug - a bug would have occurred also in other cases.

Klaus


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 8:14 am 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 14038
Location: Isleham, Cambridgeshire, UK.
klausesser wrote:
mediavets wrote:
klausesser wrote:
Related to your camera i suggest a 10,5mm Nikon fisheye for the Canon (adapter round 70.-€). This also takes 7 shots - but provides much better resolution, is a well known and supported lens - and produces less CAs @f8.

best and good luck, Klaus


That odd combination doesn't offer aperture control as I recall?


"odd combination"?? I used it myself successfully for quite a time.

For what would you need an aperture control on a fisheye?

mediavets wrote:
I think nowadays people would prefer the Sigma 10mm fisheye or the Tokina 10-17mm zoom fisheye with Canon mount


When i said Nikon 10,5mm that 1) was an example regarding the focal-length and resolution and 2nd) because i know this lens by working with it on a 20D some years ago. Excellent fisheye!

You´re experienced yourself using the Sigma 10mm FE or the Tokina 10-17mm zoom fisheye for panoramas?

Klaus


My point was simply that when you and others used the Nikon 10.5mm fisheye with a adapter on Canon cropped sensor bodies there was no alternative similar focal length fisheye with a Canon mount.

Today there are several.

I use a Nikon 10.5mm FE on a Nikon DX body.


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 8:21 am 
Offline
Member

Joined: Wed Nov 14, 2007 2:12 pm
Posts: 14038
Location: Isleham, Cambridgeshire, UK.
klausesser wrote:
thorsteinssonh wrote:
Do you know if there is a software bug tracker online for Autopano?
I take it most people commenting here are not working for Kolor on developing the Autopano interface and functionality?


I´m quite sure the issue isn´t a "bug". There are fisheye-projection models which are commonly used because most of the lenses available use the one or two certain projection-methods.

The Sunex is an extremely "exotic" lens - which needs it´s own dedicated application for undistortion and has a certain and uncommon projection-model.

This all isn´t of any problem as long as you use it for usual photography - but might cause trouble when it comes to stitching.

Whether Kolor has it in it´s database or not i don´t know. But i know that commercial/professional photographers - like myself - usually avoid using exotical gear
but rely on well proven devices.

It might figure out that the problem is this very special type of lens for which no lens-model
exists so far.

But i don´t know. What i know: its very unlikely a bug - a bug would have occurred also in other cases.

Klaus


This test report about the Sunex fisheye seems to have some relevant comments:

http://michel.thoby.free.fr/Sunex_Super ... eport.html


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 1:01 pm 
Offline
New member

Joined: Tue Aug 05, 2014 4:19 pm
Posts: 6
If you download my example above and look at the resulting hdr image, or run those images through the Autopano solver yourself, you will see that Autopano actually solves this lens fantastically well. We often get a solve error close to 1 with this lens, and the panoramas look excellent, no obvious stretching, such as bending lines or non matching edges. So, we have to guess the Autopano lens model is compatible with this lens. How else could it find a solution that even looks remotely correct?!


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 1:08 pm 
Offline
New member

Joined: Tue Aug 05, 2014 4:19 pm
Posts: 6
I removed the redundant brackets, you were right, 3 of them were repetitions. Even after fixing this issue, Autopano produces a black line. However, after improving the solve error from 2.4 down do 1.8 for this image set, we noticed that the black line becomes much less apparent. In my image processing experience (map projecting weather satellite images) I feel that this is not well defined behaviour for this kind of algorithm, and feels like an interpolation error around the wrap-around edges of the projection space. Note that all directions were sampled in those 185 deg images, and they overlap substantially, hence the reasonable solution. There should be no reason for a black line, except off course there is a special algorithm handling the wrap around case. I wouldn't know exactly how they implemented this, but they probably had to deal with the edge case.


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 1:13 pm 
Offline
New member

Joined: Tue Aug 05, 2014 4:19 pm
Posts: 6
It would be very nice to have this situation explained, because even though most people with their images don't see this issue, it may still be happening on some level, and the projection/sampling algorithm not behaving the way the software engineers intended. There can be such things as software "bugs" even though most people don't see them.

I think it would be very prudent of the Kolor team to take a look at my example and figure out why it happens. If nobody explains how and why it happens, we certainly cant call it anything but a potential bug. I worry that guessing is not sufficiently good methodology for dealing with potential bugs in a complicated piece of software such as this.


Top
 Profile  
 
PostPosted: Wed Aug 06, 2014 3:09 pm 
Offline
Member

Joined: Mon May 22, 2006 12:18 am
Posts: 7804
Location: Duesseldorf, Germany
thorsteinssonh wrote:
It would be very nice to have this situation explained, because even though most people with their images don't see this issue, it may still be happening on some level, and the projection/sampling algorithm not behaving the way the software engineers intended. There can be such things as software "bugs" even though most people don't see them.

I think it would be very prudent of the Kolor team to take a look at my example and figure out why it happens. If nobody explains how and why it happens, we certainly cant call it anything but a potential bug. I worry that guessing is not sufficiently good methodology for dealing with potential bugs in a complicated piece of software such as this.



Be assured: dealing with panos in the gigapixel range NO issue remains unseen . . . . =D :cool:

No - a possible explanation which comes to my mind: you need to draw a circle for cropping the image-content - the image-content which this lens provides is VERY small.

This can cause a hardly visible 1-pixel gap. In the rendering you definitely will see this 1-pixel gap!+
I suggest to set the window to fullscreen when setting the circle.

btw.: did you cut ALL brackets identically? Bacause having only ONE image-circle left with a non-precise cutting the circle most likely also can cause the issue.

Again: i´m very sure that this issue isn´t a "bug" - if it would be a bug it must have happen also with other stitches. But there wasn´t reports about that at all.

Besides: please do not think others are just too stupid not to see such issues "with their images" . . . . You´re just speculating without having a reasonable
idea of what you´re talking about.

Klaus

I suggest to try: put the images into Photomatix and export them as .exr or .hdr - without processing (mapping/fusioning) them of course - then you have only ONE Image per position to stitch.
This eliminates posible issues caused by layering 5 imgages per position. That would indicate whether the issue is on one of the layers or not.


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 17 posts ] 

All times are UTC + 1 hour


Who is online

Users browsing this forum: No registered users and 2 guests


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