 |
|
|
|
|
|
 |
| ||||||||
 |
 
|
|
|
|
|
|
|
|
|
| ||||||||
|
|
| 
| User list | Rules | You are not logged in.
The documentation provided with APG, both written and video tutorials, looks like of very little help in some cases!
For example: let's speak about the Mask Tool.
It doesn't tell actually "HOW" the mask tool works... it just tells the effect, or, if you prefer, the corresponding feature, that is, what that symbol does BUT it doesn't tell "HOW" to apply it... in the end it's just a matter of trials and errors. There's no logic, no criteria behind its use!
THERE'S NO METHOD!
WHEN I need a red or green mark, WHY should I use it? I mean, they don't tell HOW the mask tool does the mask! "WHAT" is the criteria? Hue? Saturation? Luminosity? Contrast? Pattern? Shape? A Combination Of? HOW the mask tool take the decision to mask!
This is undocumented! Only at Kolor knows... let's hope they will tell! 
The use of the Masking Tool, without this fundamental information, it's just a matter of trials and errors... that is a huge waist of time!
I guess we can have the same experience on other features... just different "HOWs" and "WHATs"...
Regards,
Marco
Offline
There is not much logic behind it.
Assuming you are in the editor already, first press the preview button
When preview is ready, check for things you do not want to see in endresult. Select the marker tool and then the red marker and using the scrollwheel of your mouse go to the photo in which the unwanted part is. Place a red marker on it or place a few if it was a big object that showed.
Press review again and see if anything unwanted is left.
When working with brackets you can use the red marker that looks fatter, it will function as if a marker is placed in each photo of the bracket.
Using red markers I can quickly and completely remove any tripod parts, although thanks to my nadir-shooting-tricks there is not much I have to do.
Offline
Hans,
you're just confirming my findings. Most of the features in APG are all about TRIALS AND ERRORS!
No logic, no method. At all.
Trials And Errors is only a (method) way to waist time.
Regards,
Marco
(Please, how can I make use of the "strikethrough" character in this Forum? The Button for the corresponding code doesn't work...)
Last edited by marco.lanciani (2012-12-26 00:15:25)
Offline
DrSlony wrote:
to have the mechanics of tools such as the mask tool explained.
already a small note about how it works -distance based, contrast based, brightness based- could help a lot.
greetings from germany
Chris
strikethroughtest
generates <sup> ... </sup>, but it should be <strike> ... </strike>, should be done easily.
Offline
Marco,
Maybe my logic is a bit different then yours. I use the method described all the time, for my own panorama's and for other photographers succesfully.
marco.lanciani wrote:
Hans,
you're just confirming my findings. Most of the features in APG are all about TRIALS AND ERRORS!
No logic, no method. At all.
Trials And Errors is only a (method) way to waist time.
Regards,
Marco
(Please, how can I make use of the "strikethrough" character in this Forum? The Button for the corresponding code doesn't work...)
Offline
Mask tutorials are in progress.
This feature is based on our antighost algorithm. This algorithm tries to find the cutting way where pictures are the most similar (where pixels of different pictures are idealy equals). In maths terms, this is done thanks min-cut/max-flow algorithms.
When you add a green marker, some pixels around this position are obligatory kept. So, antighost algorithm will try to conserve the whole "object" around marker (this is the inverse for red markers)
It's intuitively hard to understand that markers are not working like simple gum. This tool is very powerful if "object" is well determined. For example, it is hard to keep a person which not entirely present on one picture because ideal transition with other pictures does not exist. In these cases where "best cutting way" is more subjective, we think tools like photoshop are more appropriated.
We think this tool is very powerful in most of antighost cases even if I'm agree that usage is not always intuitive.
We hope to well explain different cases in upcoming tutorials.
Offline
renan wrote:
We think this tool is very powerful in most of antighost cases even if I'm agree that usage is not always intuitive.
We hope to well explain different cases in upcoming tutorials.
Hi Renan!
First of all i wish you and the gang a happy, healthy and successful new year! 
I agree: your masking tool is great. I prefer it over PZGui´s masking tool - because it blends softer and works intuitive. But we really need to know when it´s preferable to set more than one marker on an object or whether it´s preferable to set "keep it" markers to adjacent images which show the background without the object.
Also it would be vital - like in the optimizer - being able to set a range/radius the blending-/CP-searching zone reaches. This way it would be better guessable where to set a marker. I usually set two or more markers to a person as example to have her/him deleted. Most of the tome it works - but it´s hardly reliable and no way precise.
But IF it works it´s simply great!!
best, Klaus
Last edited by klausesser (2012-12-31 12:26:52)
Offline
renan wrote:
Mask tutorials are in progress.
This feature is based on our antighost algorithm. This algorithm tries to find the cutting way where pictures are the most similar (where pixels of different pictures are idealy equals). In maths terms, this is done thanks min-cut/max-flow algorithms.
When you add a green marker, some pixels around this position are obligatory kept. So, antighost algorithm will try to conserve the whole "object" around marker (this is the inverse for red markers)
It's intuitively hard to understand that markers are not working like simple gum. This tool is very powerful if "object" is well determined. For example, it is hard to keep a person which not entirely present on one picture because ideal transition with other pictures does not exist. In these cases where "best cutting way" is more subjective, we think tools like photoshop are more appropriated.
We think this tool is very powerful in most of antighost cases even if I'm agree that usage is not always intuitive.
We hope to well explain different cases in upcoming tutorials.
thanks for explanation. So if take each image several times with some time delay, but with complete identical values (shutter,focal,wb,aperture) and I do group them in image stacks inside autopano (not with intention to hdr them) than I do have 100% overlapping on each image position and in consequence the stack masking tool should be able to delete all ghosts? (currently it fails in those cases where no overlapping to neigbor images does exist)
If this does work than it would be wonderful to add some sort of denoising algorithm too so the multiple images have even more advantages 
Goerg
Last edited by gkaefer (2012-12-31 13:21:19)
Offline
Renan, it is working like a charm. It is like you explain it, not much more to say about it.
A little tip, use the green more to the middle of a photo and the reds more on the outside. Many lenses are better in the center then at the outside.
Offline
gkaefer wrote:
So if take each image several times with some time delay, but with complete identical values (shutter,focal,wb,aperture) and I do group them in image stacks inside autopano (not with intention to hdr them) than I do have 100% overlapping on each image position and in consequence the stack masking tool should be able to delete all ghosts? (currently it fails in those cases where no overlapping to neigbor images does exist)
This works. But it takes additional time for the shooting and can become somewhat confusing with lots of people. In the case you have many moving objects there is no way around using NO bracketing and having short exp. time.
Sometimes i use the speed-mode - but that´s of no help with moving objects at all and you would need to 1) use extremely short exp. times=extremely bright light and 2) you would need to match the velocity of the spinning head and the speed of the moving objects for NOT having multiple images containing the same objects.
In the case you have moving objects a very slow frequency of shots definitely is preferable because the objects have moved a greater distance then and can be edited very much easier - if that´s necessary at all.
Unexperienced users sometimes believe the speed-mode might be preferable in crowded places - that´s definitely not the case.
best, Klaus
Last edited by klausesser (2012-12-31 14:25:16)
Offline
gkaefer wrote:
.......
So if take each image several times with some time delay, but with complete identical values (shutter,focal,wb,aperture) and I do group them in image stacks inside autopano (not with intention to hdr them) than I do have 100% overlapping on each image position and in consequence the stack masking tool should be able to delete all ghosts? (currently it fails in those cases where no overlapping to neigbor images does exist)
If this does work than it would be wonderful to add some sort of denoising algorithm too so the multiple images have even more advantages
Goerg
Hmmmmm, in situations with many people walking around we often shoot the same stack a few times. Your idea seems to suggest to join these stacks together to see how that works out, or just place the reference photos of these stacks together into one stack. Will try to do some testing here, intereseting.
Offline
HansKeesom wrote:
Your idea seems to suggest to join these stacks together to see how that works out, or just place the reference photos of these stacks together into one stack. Will try to do some testing here, intereseting.
The reason is that moving people don´t cut themselves multiple times. Doing more shots most likely they move a greater distance between shots and this way can be erased and replaced by the background much better.
Btw.: the great misundersanding in using the speed-mode on robotic heads - which means a fast spinning non-stop moving while firing the camera also nonstop - is that in this mode people don´t move on far enough between shots. This way you have multiple overlaps from a person without seeing the person´s background alone. That makes masking the person impossible resp. at least extremely difficult.
So using the speed-mode in surroundings with fast moving objects is one of the wrongest things one can do panorama-wise . . . Much better is the way Georg describes - bigger effort, butdefinitely more reliable.
In the example of the speed-mode on the Rodeon-site there is a car moving in the same direction the Rodeon moves - and so the camera sees the car in a number of shots moving in parallel to the camera - can you imagine how THAT looks like in the stitch? (i know the result . . )
I had a long talk with Susanne Clauss - who is visible in the video - about the speed-mode (and other items of course): we´re in no way apart in our judgement about it :
Under SOME aspects it´s a good feature. But too many people completely overestimate the usability of it and imagine to use it in situations where it simply would be nonsense to do.
best, Klaus
Offline
using fisheyes the situation is clear. less to no need to take the additional effort into consideration... moving objects are less a problem.
but if I do a >35mm than the FOV is so small that ghosts cover a huge part of the image (sharp or nonsharp). so autopanos masking cant work on parts of the image where no overlapping does exist to neigbor images - in special the middle center part. so here using this kind of stacking could be solved. but currently cant be combinded with the normal stacking of exposures.
Georg
Last edited by gkaefer (2013-01-01 18:34:38)
Offline
Some explanations about ghost and stacks.
Mask markers are used by anti ghost algorithm (named "Cutting -> Ghost" in render parameters). The result is binary, when one pixel is choosed, other pictures information are rejected at this location. So, if we apply this algorithm on stacks, exposure fusion can't be done because only one pixel per location is kept.
It's why cutting algrithm is applied by layers only (on same exposure level in standard cases) and not between all pictures.
So, for "deghosting" a same location (a stack), you have to break the stack to have one layer only. But in this case, fusion can not be applied.
"HDR ghost" parameter try to reduce ghost effects between layers. It's named "HDR Ghost" because it's working well when pictures have the same exposition or when expostions are compensated (with HDR color correction). But the choice is not binary and result is not like "Cutting -> Ghost".
Offline
renan wrote:
Some explanations about ghost and stacks.
Mask markers are used by anti ghost algorithm (named "Cutting -> Ghost" in render parameters). The result is binary, when one pixel is choosed, other pictures information are rejected at this location. So, if we apply this algorithm on stacks, exposure fusion can't be done because only one pixel per location is kept.
It's why cutting algrithm is applied by layers only (on same exposure level in standard cases) and not between all pictures.
So, for "deghosting" a same location (a stack), you have to break the stack to have one layer only. But in this case, fusion can not be applied.
"HDR ghost" parameter try to reduce ghost effects between layers. It's named "HDR Ghost" because it's working well when pictures have the same exposition or when expostions are compensated (with HDR color correction). But the choice is not binary and result is not like "Cutting -> Ghost".
if I use 5 images per image position + all images are equally shot with same values for focal, wb, aperture, shutter + I "group" them in autopano per N=5 files:
1. its obvious that I wont do fusion (for HDR or tonemapping)
2. the intention for doing this only is to have 5 images per imageposition where I can aplly the existing masking tool to 100% of the image area (and not only to the overlapping areas to the neigbor images when using only 1 image per position or if using image groups/stacks for doing fusion with intention for HDR/tonemapped result.
so this should work with current version.
(it would be genious now if autopano does recognize such image groups and in such cases a new feature/button gets active which now allows denoising the image group resulting in a single image and the so denoised images now can be stiched to a pano...)
Georg
Last edited by gkaefer (2013-01-02 12:28:03)
Offline
renan wrote:
It's why cutting algrithm is applied by layers only (on same exposure level in standard cases) and not between all pictures.
So, for "deghosting" a same location (a stack), you have to break the stack to have one layer only. But in this case, fusion can not be applied.
Hi Renan!
To avoid misunderstandings: that means to use the ONE-layer (no stacks) method from the beginning to use it?
renan wrote:
when pictures have the same exposition or when expostions are compensated (with HDR color correction). But the choice is not binary and result is not like "Cutting -> Ghost".
What´s the meaning of "exposition" here exactly?
best, Klaus
Offline
klausesser wrote:
To avoid misunderstandings: that means to use the ONE-layer (no stacks) method from the beginning to use it?
Yes, that means layer only. With one layer anti ghost will consider all pictures together.
klausesser wrote:
What´s the meaning of "exposition" here exactly?
Actually, it's pictures of same layer. Layers are often automatically created on EV differences. It's not rigorously the same exposition but the same EV difference. But anyway, you can create layers as you want (not only based on stack) and you have to retain that anti ghost is applied between pictures of same layer.
gkaefer wrote:
the intention for doing this only is to have 5 images per imageposition where I can aplly the existing masking tool to 100%
If I well understand, just put all pictures in same layer (or do not create stack in group view) and it must working as you want (if the solution exists).
Offline
renan wrote:
klausesser wrote:
To avoid misunderstandings: that means to use the ONE-layer (no stacks) method from the beginning to use it?
Yes, that means layer only. With one layer anti ghost will consider all pictures together.
klausesser wrote:
What´s the meaning of "exposition" here exactly?
Actually, it's pictures of same layer. Layers are often automatically created on EV differences. It's not rigorously the same exposition but the same EV difference. But anyway, you can create layers as you want (not only based on stack) and you have to retain that anti ghost is applied between pictures of same layer.
gkaefer wrote:
the intention for doing this only is to have 5 images per imageposition where I can aplly the existing masking tool to 100%
If I well understand, just put all pictures in same layer (or do not create stack in group view) and it must working as you want (if the solution exists).
Hi Renan!
Thank you for explaining!
best, Klaus
Offline
Hi Renan,
thanks for your explanation. It helps, by the way...
Basically, you're asking a (min-cut/max-flow) algorithm to guess the selection by guessing the relevant pixels!
Looks like shooting at flies with cannons! Maybe it’s not a case if your red/green marks looks like a target.
A double guess is nothing close to predictable: you can either hit the target or miss it... and I predict long life to that fly.
I'm sure you agree that without a selection there can be any mask nor deghosting...
Photomatix has a nice deghosting option. It's not a total guess: it’s applied locally, inside a precise selection.
Your idea is good, and looks very close to that of Photomatix, but needs some work: just a better selection tool, like a lazo or a brush with variable radius.
Masking (cutting), ghosting (HDR Ghost), stacks... as you explain, this is a complex topic but I feel like it’s the CORE topic around which I took the decision on, to go with APG as my main App. That’s why either more detailed explanations or more intuitive tools, as you do pointed out, are needed.
Regards,
Offline
Christian Stüben wrote:
strikethroughtest
generates <sup> ... </sup>, but it should be <strike> ... </strike>, should be done easily.
Hi Chris, I tried your suggestion, I even searched the web... doesn't work. There might be some limitation within the forum...
Offline
