I have played a little with 3Denlive, and followed the discussion on the 3D Spider-Man.
I am not fully on grips of the application but I feel it is extremely interesting to produce videos out of a sequence of stereo images, including zoom-in/out effects, and overlaying two or more stereo images.
In my opinion 3Denlive as a video editor would work best if the input SBS images are already well aligned for the main angular parameters (visual axis rotation, h-perspective, v-perspective, and nominal focus correction). The only alignment parameter that makes sense for zooming in-out is the h-disparity (displacement in the X direction).
As an example of this take one stereo image that has a 3.3% horizontal disparity. If we make a video sequence that ends in 5x zoom-in, the disparity would become 16.5% which is obviously too much. This is were it makes sense to displace in the X direction trying to reduce the disparity in the zoomed-in area, or possibly masking some area of the image that creates excessive disparity.

Regards
   Antonio

Le 26/11/2020 à 11:14, Antonio F.G. via groups.io a écrit :

I have played a little with 3Denlive, and followed the discussion on the 3D Spider-Man.
I am not fully on grips of the application but I feel it is extremely interesting to produce videos out of a sequence of stereo images, including zoom-in/out effects, and overlaying two or more stereo images.
In my opinion 3Denlive as a video editor would work best if the input SBS images are already well aligned for the main angular parameters (visual axis rotation, h-perspective, v-perspective, and nominal focus correction). The only alignment parameter that makes sense for zooming in-out is the h-disparity (displacement in the X direction).
As an example of this take one stereo image that has a 3.3% horizontal disparity. If we make a video sequence that ends in 5x zoom-in, the disparity would become 16.5% which is obviously too much. This is were it makes sense to displace in the X direction trying to reduce the disparity in the zoomed-in area, or possibly masking some area of the image that creates excessive disparity.

Exactly !

As shown in this tutorial about how to use keyframes, you can control all alignment and placement/zoom parameters to make them adapted at each time position of the video.

http://3denlive.com/doc/EN/KeyframedAnimation.php

You are speaking about horizontal disparity. But, on some stereos, you may have a well-aligned whole image having some local deformations. When zooming on these details, you may need to completely re-align the zoomed sub-part. This is here where the "Automatic alignment" button takes a great part of its interest!

For example, in the Spider-Man case, you may have a well-align whole video, but it's very hard to get it properly with a well-aligned Spider-Man (in fact, the background seems to be a simple flat wall-paper!). You would certainly like to have a better alignment on it when zooming.

In the tutorial, see the first screen copy point 7, and third screen copy point 3 and 4.

Your explanations may be a bit confusing because there are two X tuning:

- the horizontal stereo shift in the Alignment configuration

- the horizontal position in the Placement configuration

In the same idea, there are two parameters that may change the size aspect:

- the horizontal and vertical adaptation in the Alignment configuration

- the zoom factor in the Placement configuration

;-)

On Thu, Nov 26, 2020 at 04:41 AM, Etienne Monneret (Perso) wrote:

You are speaking about horizontal disparity. But, on some stereos, you
may have a well-aligned whole image having some local deformations.

If they have local deformations they are not very well aligned.

When zooming on these details, you may need to completely re-align the
zoomed sub-part. This is here where the "Automatic alignment" button takes a
great part of its interest!

OK, but it only makes sense if the original SBS are poorly aligned

For example, in the Spider-Man case, you may have a well-align whole
video, but it's very hard to get it properly with a well-aligned
Spider-Man (in fact, the background seems to be a simple flat
wall-paper!).

Spider-Man is a composition of two stereo pairs: the background and the
Spider-Man proper. If each pair were perfectly aligned in the origin, then
the composition of the man overlaying the background would be perfectly
aligned as well.
The composition would surely require adjusting the depth of the Spider-Man
respect the background, because it would be bad if it appeared behind the
background, and it would also be bad if it appeared too much in front. This
is where I believe your "X" control in "Alignement" makes sense. They also
make sense the "X", "Y" and "D" of "Placement" because they do not spoil the
alignment, they only control the position and the zoom of the complete overlay.

But frankly, I would not use the "Y", "L", "H" and "R" of "Alignement"
because they spoil the vertical alignment, and it is very difficult to assess
the error visually.

Your explanations may be a bit confusing because there are two X tuning:

- the horizontal stereo shift in the Alignment configuration
- the horizontal position in the Placement configuration

OK, I understand the first for adjusting the depth of an image respect the
window, and the second for displacing both stereo sides at the same time
(I guess for placing overlays)

Regards
   Antonio

On Thu, Nov 26, 2020 at 04:41 AM, Etienne Monneret (Perso) wrote:

You are speaking about horizontal disparity. But, on some stereos, you
may have a well-aligned whole image having some local deformations.

If they have local deformations they are not very well aligned.

 

It depends on the quality of the source.

If this is a cha-cha, you may have important local deformations.

With stereo dedicated hardware, you may have time shift, lens differences, bad aligned mirrors, focal or zoom differences.

There are many reasons why an original may not be perfect!

When zooming on these details, you may need to completely re-align the
zoomed sub-part. This is here where the "Automatic alignment" button takes a
great part of its interest!

OK, but it only makes sense if the original SBS are poorly aligned

 

Or badly shots.

For example, in the Spider-Man case, you may have a well-align whole
video, but it's very hard to get it properly with a well-aligned
Spider-Man (in fact, the background seems to be a simple flat
wall-paper!).

Spider-Man is a composition of two stereo pairs: the background and the
Spider-Man proper. If each pair were perfectly aligned in the origin, then
the composition of the man overlaying the background would be perfectly
aligned as well.

 

The background is a flat wallpaper, and yes, Spider-Man itself is not well aligned.

The whole image can be "at-best" aligned, but it's impossible to get it "perfectly-aligned" without doing a complex morphing.

The composition would surely require adjusting the depth of the Spider-Man
respect the background, because it would be bad if it appeared behind the
background, and it would also be bad if it appeared too much in front.

 

The original is "at it is", and you can't change the relative stereo position of the foreground with the background without doing a morphing. Whatever you do with simple transformations (translation, rotation, scale, width, height, etc), Spider-Man will always stay in front of the background.

This
is where I believe your "X" control in "Alignement" makes sense. They also
make sense the "X", "Y" and "D" of "Placement" because they do not spoil the
alignment, they only control the position and the zoom of the complete overlay.

But frankly, I would not use the "Y", "L", "H" and "R" of "Alignement"
because they spoil the vertical alignment, and it is very difficult to assess
the error visually.

 

It's completely dependant on the source.

Of course, if you have a perfect stereo original, you won't have to use them.

In fact, there is one case where you would need it, even with a perfect hardware: with an hyper-stereo near first plan. When shooting from too near positions for the first plan regarding the space between lens (too large for that distance), parts of the image will be nearer on one side than the other side, and when say nearer say larger. You will then be obliged to balance the alignment to keep with these larger/smaller parts, as we did with the Spider-Man example.

;-)

On Fri, Nov 27, 2020 at 05:06 AM, Etienne Monneret (Perso) wrote:

If they have local deformations they are not very well aligned.

It depends on the quality of the source.

If this is a cha-cha, you may have important local deformations.

With stereo dedicated hardware, you may have time shift, lens differences, bad aligned mirrors, focal or zoom differences.

There are many reasons why an original may not be perfect!

I view 3De as a very promising application to generate video from stereo stills. Not so much for stereo aligning because I trust better my own STMANI3, which does a better alignment job than SPM, if only because with STMANI3 I can control the placement of the matching points and also have better control of the alignment variables than SPM provides.
All my SBS are (almost:-) perfectly aligned in the whole scene, so I intend to use them as a source for your very promising 3De. This why I believe that only X alignment (stereo depth) may be relevant for the video zooming and overlays.

Granted, it is not possible to align moving scenes not shot synchronously. Perhaps here does make sense the 3De feature of aligning local areas. Also STMANI3 can not (yet) align images with gross lens distortions, though there are ways to solve this. Any other miss-shooting can be perfectly aligned with an appropriate perspective transformation.

 

In fact, there is one case where you would need it, even with a perfect hardware: with an hyper-stereo near first plan. When shooting from too near positions for the first plan regarding the space between lens (too large for that distance), parts of the image will be nearer on one side than the other side, and when say nearer say larger. You will then be obliged to balance the alignment to keep with these larger/smaller parts, as we did with the Spider-Man example.

Yes, this sort of too-near objects produce retinal rivalry. My recipe is masking them out. STMANI3 has feature for this as shows this example from the STMANI3 user's manual where an unwanted rock appears in the right image:


Regards
   Antonio

In fact, there is one case where you would need it, even with a perfect hardware: with an hyper-stereo near first plan. When shooting from too near positions for the first plan regarding the space between lens (too large for that distance), parts of the image will be nearer on one side than the other side, and when say nearer say larger. You will then be obliged to balance the alignment to keep with these larger/smaller parts, as we did with the Spider-Man example.

Yes, this sort of too-near objects produce retinal rivalry. My recipe is masking them out. STMANI3 has feature for this as shows this example from the STMANI3 user's manual where an unwanted rock appears in the right image:

 

I was speaking about hyper-stereo objects in near foreground. But, this rock is not a stereo at all.

:)

Antonio,

On Sat, Nov 28, 2020 at 08:29 AM, robert mcafee wrote:

Does your STMANI program provide masking capability

Yes, it is part of the program. It is explained in this part of the user's manual:
http://afgalaz.es/stmani3/en/angular_alignment.html#masks

with the stereomasken program?

I am just looking the stereomasken site. I never used but it seems a very fancy stereo mask tool, with rounded splines, different depths along the margin et al.
The masks of STMANI3 are just for cutting out selected part of the images. The are open polygonals with every vertex at the same depth as the margins.

Regards
   Antonio