3D Pop-Left (or Right)


 

So... a 3D pop-left (or right) is working without the neck-breaking agony of creating and viewing a 3D pop-down.
The 3D camera angle is 45 degrees (left or right), and you have to view them with that same angle.
 
Hope this helps.

Parallel view.


 

That is how it looks in virtual reality (and in reality), so that is not the image that you rotate. This is the image you rotate 45 degrees:


Etienne Monneret (Perso)
 

Le 24/04/2021 à 03:10, Roger Maddy a écrit :
That is how it looks in virtual reality (and in reality), so that is not the image that you rotate. This is the image you rotate 45 degrees:

Yes!

Your image shows a rectangle.

When viewed in you VR, it becomes a trapezoid.

When you will shoot such an object in reality (the rectangle being for example a table), you will also get a trapezoid, like in your VR.

To get the good image in order to have the phatogram, you will then be obliged to reverse the perspective effect, transform the shot trapzoid in the original rectangle.

The good news is that this perspective inversion can be done within your VR by applying an inverse rotation: -45° in place of +45°:

  1. shoot your real scene with a +45° angle, your table should be a trapezoid.
  2. display this image in your VR with a -45° angle (this will reverse the perspective obtained at step 1), your table should now be a rectangle.
  3. shoot the VR result, this will be your phantogram
  4. display this phantogram by applying a +45° angle (in fact +90° comparing to the VR shooting time at step 2-3), your table is again a trapezoid, like at step 1.

;-)



Etienne Monneret (Perso)
 

Le 24/04/2021 à 05:52, Etienne Monneret (Perso) a écrit :

The good news is that this perspective inversion can be done within your VR by applying an inverse rotation: -45° in place of +45°:

  1. shoot your real scene with a +45° angle, your table should be a trapezoid.
  2. display this image in your VR with a -45° angle (this will reverse the perspective obtained at step 1), your table should now be a rectangle.
  3. shoot the VR result, this will be your phantogram
  4. display this phantogram by applying a +45° angle (in fact +90° comparing to the VR shooting time at step 2-3), your table is again a trapezoid, like at step 1.

In step 4, the viewing angle becomes +45°. But, as you are also looking at a +45° angle (for example looking down to a table), the angle to apply to your phantogram obtain in step 3 in the VR scene is +90° (it should be horizontal), this means that the table is rotated +135° comparing to the VR shooting time at step 2-3.



Antonio F.G.
 
Edited

I am just starting to understand this phantogram trick as a pop-up effect (or pop-down if I could place the image in the ceiling:-). In both cases a vertical perspective (vertical keystone) could be applied. I have worked a little with perspective alignment, and I can say that vertical perspective does not worsen the stereo alignment, i.e. you can modify the vertical keystone of any aligned stereo pair at will, and be sure the image while distorted will still keep aligned.

But you are talking of pop-right, pop-left. This is a very different thing, because horizontal perspective (h. keystone) does affect the stereo alignment. If you apply an H. perspective to an aligned pair you will worsen the alignment. In your example of the horizontal cone, what if you had not cropped out all the background of the original images?. I guess it would be un-viewable because of vertical disparity in the background. The image that shows the cone with a 45˚ h. perspective is viewable only because there is no background.

Regards
    Antonio


On Fri, Apr 23, 2021 at 07:45 PM, Roger Maddy wrote:
So... a 3D pop-left (or right) is working without the neck-breaking agony of creating and viewing a 3D pop-down.
The 3D camera angle is 45 degrees (left or right), and you have to view them with that same angle.
 
Hope this helps.

Parallel view.


Etienne Monneret (Perso)
 

There is no link between the stereo keystone problem and the phantogram transformation.

It's only a question of what should be print according to the viewing position, nothing more.

Common stereo are made to be viewed as the cameras shot them, thus your viewing line must be perpendicular to the plan of the picture. If you look at it looking down at a 45°, to see what the cameras was shooting, you need to keep it perpendicular (red line).

Now, if you want to look at it put horizontally on a table, to be able to see what the camera was shooting, you need to make a perspective projection (blue line). This is what is doing the anamorphosis (reversing the original perspective effect).

You can do this projection in any kind of direction!



Le 24/04/2021 à 14:56, Antonio F.G. via groups.io a écrit :

[Edited Message Follows]

I am just starting to understand this phantogram trick as a pop-up effect (or pop-down if I could place the image in the ceiling:-). In both cases a vertical perspective (vertical keystone) could be applied. I have worked a little with perspective alignment, and I can say that vertical perspective does not worsen the stereo alignment, i.e. you can modify the vertical keystone of any aligned stereo pair at will, and be sure the image while distorted will still keep aligned.

But you are talking of pop-right, pop-left. This is a very different thing, because horizontal perspective (h. keystone) does affect the stereo alignment. If you apply an H. perspective to an aligned pair you will worsen the alignment. In your example of the horizontal cone, what if you had not cropped out all the background of the original images?. I guess it would be un-viewable because of vertical disparity in the background. The image that shows the cone with a 45˚ h. perspective is viewable only because there is no background.

Regards
    Antonio


On Fri, Apr 23, 2021 at 07:45 PM, Roger Maddy wrote:
So... a 3D pop-left (or right) is working without the neck-breaking agony of creating and viewing a 3D pop-down.
The 3D camera angle is 45 degrees (left or right), and you have to view them with that same angle.
 
Hope this helps.

Parallel view.



Etienne Monneret (Perso)
 

Note that  the anamorphosis is not the same for left and right views: there is a small horizontal shift according to the stereo-base. The different perspective projections (from red plan to blue plan) should set the parallax to zero for the whole ground plan (reference square).

Le 24/04/2021 à 15:09, Etienne Monneret (Perso) a écrit :
There is no link between the stereo keystone problem and the phantogram transformation.

It's only a question of what should be print according to the viewing position, nothing more.

Common stereo are made to be viewed as the cameras shot them, thus your viewing line must be perpendicular to the plan of the picture. If you look at it looking down at a 45°, to see what the cameras was shooting, you need to keep it perpendicular (red line).

Now, if you want to look at it put horizontally on a table, to be able to see what the camera was shooting, you need to make a perspective projection (blue line). This is what is doing the anamorphosis (reversing the original perspective effect).

You can do this projection in any kind of direction!



Le 24/04/2021 à 14:56, Antonio F.G. via groups.io a écrit :

[Edited Message Follows]

I am just starting to understand this phantogram trick as a pop-up effect (or pop-down if I could place the image in the ceiling:-). In both cases a vertical perspective (vertical keystone) could be applied. I have worked a little with perspective alignment, and I can say that vertical perspective does not worsen the stereo alignment, i.e. you can modify the vertical keystone of any aligned stereo pair at will, and be sure the image while distorted will still keep aligned.

But you are talking of pop-right, pop-left. This is a very different thing, because horizontal perspective (h. keystone) does affect the stereo alignment. If you apply an H. perspective to an aligned pair you will worsen the alignment. In your example of the horizontal cone, what if you had not cropped out all the background of the original images?. I guess it would be un-viewable because of vertical disparity in the background. The image that shows the cone with a 45˚ h. perspective is viewable only because there is no background.

Regards
    Antonio


On Fri, Apr 23, 2021 at 07:45 PM, Roger Maddy wrote:
So... a 3D pop-left (or right) is working without the neck-breaking agony of creating and viewing a 3D pop-down.
The 3D camera angle is 45 degrees (left or right), and you have to view them with that same angle.
 
Hope this helps.

Parallel view.




JackDesBwa|3D
 

I agree that it should be possible, and wanted to see how it looks like in reality.

Then I imagined a head popping from a hole in a wall, exported a 3d image and transformed it quickly (alignment is not perfect) into a phantogram. The 3D scene does not look like I wanted, but the effect works.

I attached the test images if you want to see them. [parallel versions]
The image named 'scene' is the export from the modeling software, which also helps to see how it looks like with a regular camera.
The other image named 'result' is the phantogram to see from the side (image at 45° on your left, facing to your right).
view.png

It is not as pleasing to view as regular phantograms in my opinion, because turning the head down to view a figurine is quite natural, but in this case with the image on the left, I want to adjust my position to come back in front of it.

JackDesBwa


Antonio F.G.
 

Granted, the phantogram the transformation may be in any direction.
But if it is stereo there are other difficulties. Most important, there should not be vertical disparity. This is where I doubt that you can apply perspective transformation in any direction. On my view only the vertical perspective does not worsen the stereo alignment (vertical disparity).

Regards
     Antonio


Etienne Monneret (Perso)
 

Le 25/04/2021 à 00:08, Antonio F.G. via groups.io a écrit :
But if it is stereo there are other difficulties. Most important, there should not be vertical disparity. This is where I doubt that you can apply perspective transformation in any direction. On my view only the vertical perspective does not worsen the stereo alignment (vertical disparity).

Phantogram does not create stereo disparity problem.

When you do the projection, of course the result, seen as a direct orthogonal viewing, may show deformations like the well-known stereo keystone problem, especially if this is a vertical phantogram.

In the Jack's example, it can already be seen on the place where there is the larger parallax:

BUT, the phantogram must not be seen with a perpendicular viewing, like an ordinary stereo. It must be seen with a 45° angle. In this way, the perspective of the phantogram itself do a reverse disparity correction.

By construction, your eyes always see exactly the same thing at 45° on the phantogram, that what was seen at perpendicularly on the original stereo before the phantogram projection.

It's understandable on my drawing: point to point in the image, each eye see EXACTLY the same thing. If you do not have bad disparity on the original image (red plan), you won't see any disparity in the projected image (blue plan), even if this projected image shows disparities when seen perpendicularly (it's not intended to be seen perpendicularly).



Etienne Monneret (Perso)
 

Le 24/04/2021 à 23:18, JackDesBwa|3D a écrit :
Then I imagined a head popping from a hole in a wall, exported a 3d image and transformed it quickly (alignment is not perfect) into a phantogram.
Why "alignment is not perfect"? It seems quite good for me. Small vertical disparities in the larger parallax are normal (see my other message).


JackDesBwa|3D
 

Why "alignment is not perfect"? It seems quite good for me.

It is not perfect in the sense that it is not computed as the exact projection, that could be awaited from a modeled scene.

I used the perspective tool of the Gimp to go from the rendered image to the phantogram image, by placing the grid manually over the square.
This results in small inaccuracies, such as the plane not being exactly at zero parallax for example. But the difference with ideal is indeed small, except perhaps in terms of resolution.

JackDesBwa


Antonio F.G.
 

It sounds correct: the horizontal perspective induces vertical disparity errors but they are cancelled when viewed from an angle exactly equal to that of the perspective.

That means the pop-right, pop-left phantograms can indeed work, but may be critical in regard to the viewer position. If the viewer looks from a non optimal angle, then vertical disparity would appear.

In contrast I think pop-up, pop-down phantograms should not produce any vertical disparity regardless the angle they are viewed from. By modifying the angle they will show funny perspective effects, but at least no head-aches:-).

All this sounds very interesting. For Stmani3 I have to think if I could set a condition of zero-disparity for a set of points. Perhaps only three points to define a plane.

Regards
    Antonio


Etienne Monneret (Perso)
 

Le 25/04/2021 à 23:24, Antonio F.G. via groups.io a écrit :
All this sounds very interesting. For Stmani3 I have to think if I could set a condition of zero-disparity for a set of points. Perhaps only three points to define a plane.
The best is to shoot the scene with a reference plan shape somewhere (ideally a square, as you said, you only need 3 points), this shape being horizontal on the reference (ground) level, and then to change each shot view with a perspective transformation to get back exactly the original shape with zero parallax. So, in your process, it would be nice to load 3 images: left, right, and original shape alone. Of course, it would often be nice to be able to remove the shape on the final stereo, or use a shape that could be nicely integrated in the scene (like my chess board).


robert mcafee
 

Barry Rothstein has published several books of phantom rams which provide information on making your own. He also has information on his websites. Barry did a live tutorial at the NYSA stereo club Zoom meeting this past Saturday. He showed his process with photoshop and using StereoPhoto Maker. 




On Monday, April 26, 2021, 1:35 AM, Etienne Monneret (Perso) <ab3d@...> wrote:

Le 25/04/2021 à 23:24, Antonio F.G. via groups.io a écrit :
> All this sounds very interesting. For Stmani3 I have to think if I
> could set a condition of zero-disparity for a set of points. Perhaps
> only three points to define a plane.

The best is to shoot the scene with a reference plan shape somewhere
(ideally a square, as you said, you only need 3 points), this shape
being horizontal on the reference (ground) level, and then to change
each shot view with a perspective transformation to get back exactly the
original shape with zero parallax. So, in your process, it would be nice
to load 3 images: left, right, and original shape alone. Of course, it
would often be nice to be able to remove the shape on the final stereo,
or use a shape that could be nicely integrated in the scene (like my
chess board).








Etienne Monneret (Perso)
 

Le 26/04/2021 à 12:22, robert mcafee via groups.io a écrit :
Barry Rothstein has published several books of phantom rams which provide information on making your own. He also has information on his websites. Barry did a live tutorial at the NYSA stereo club Zoom meeting this past Saturday. He showed his process with photoshop and using StereoPhoto Maker.
I found his tutorial on this web site:

http://3ddigitalphoto.com/

Direct link to the document:

http://3ddigitalphoto.com/phantograms.pdf

;-)


Antonio F.G.
 

I have read the tutorial of Barry Rothstein that you sent: http://3ddigitalphoto.com/ According to him this technique requires a lot of precision for shooting the photos (it almost requires screwing the tripod to the floor:-).
He gives several phantogram examples, but no one of pop-right, pop-left. The only example he gives of a vertical phantogram is shot frontally, i.e. the stereo base is parallel to the phantogram frame, as are the horizontal ones, so the effect should still be a pop-up, not a lateral pop.

I decide to make a drawing in stereo to clear my mind:
  https://stereopix.net/photo:9ON3DvIwow/
My conclusion for now is that the stereo base must be parallel to the phantogram frame. If this condition is met we can get two exact copies of the object frame by re-projecting the images in a Virtual Sensor Plane that is parallel to the plane of the object frame.
The pointing of the cameras is not important (as long as they pick the scene), because pointing misalignment is corrected in the re-projection.

I do not understand the need of three images. What is the "original shape alone" for? As far as I understand the phantogram frame is already present in the left and right images (Barry Rothstein dixit)

Regards
    Antonio


On Mon, Apr 26, 2021 at 12:35 AM, Etienne Monneret (Perso) wrote:
Le 25/04/2021 à 23:24, Antonio F.G. via groups.io a écrit :
All this sounds very interesting. For Stmani3 I have to think if I
could set a condition of zero-disparity for a set of points. Perhaps
only three points to define a plane.
The best is to shoot the scene with a reference plan shape somewhere
(ideally a square, as you said, you only need 3 points), this shape
being horizontal on the reference (ground) level, and then to change
each shot view with a perspective transformation to get back exactly the
original shape with zero parallax. So, in your process, it would be nice
to load 3 images: left, right, and original shape alone. Of course, it
would often be nice to be able to remove the shape on the final stereo,
or use a shape that could be nicely integrated in the scene (like my
chess board).


Etienne Monneret (Perso)
 

Le 27/04/2021 à 23:13, Antonio F.G. via groups.io a écrit :
He gives several phantogram examples, but no one of pop-right, pop-left. The only example he gives of a vertical phantogram is shot frontally, i.e. the stereo base is parallel to the phantogram frame, as are the horizontal ones, so the effect should still be a pop-up, not a lateral pop.
Of course, phantograms are mostly understood as pop-up anaglyphs. But, this do not prevent from making pop-right or pop-left.


I decide to make a drawing in stereo to clear my mind:
https://stereopix.net/photo:9ON3DvIwow/ <https://stereopix.net/photo:9ON3DvIwow/>
Can't figure out how to understand your drawing. At least I need some explanations.


My conclusion for now is that the stereo base must be parallel to the phantogram frame.
Not necessarily: with a pop-right or pop-left phantogram, the stereo-base is perpendicular to the phantogram frame. In fact, I think the only real constraints is that the viewing position of the phatogram must be the same as the shooting position of the cameras.


If this condition is met we can get two exact copies of the object frame by re-projecting the images in a Virtual Sensor Plane that is parallel to the plane of the object frame.
The pointing of the cameras is not important (as long as they pick the scene), because pointing misalignment is corrected in the re-projection.
Not sure to understand you. Whatever the shooting angle, the final viewing will be at the same angle.



I do not understand the need of three images.
This is the reason why I think you still miss something (see below).


What is the "original shape alone" for? As far as I understand the phantogram frame is already present in the left and right images (Barry Rothstein dixit)
The goal of the phantogram perspective projection is to restore the reference shape as it was originally when properly viewed in-front, thus exactly reversing the perspective effect obtained at the shooting time. It's not only matching both views to set a zero parallax.

If this shape is a perfect square, of course, the problem is simple: you just need to find the perspective projection restoring a perfect square on each view, with a zero parallax between them. This is the reason why most phantograms are built using a square as the reference shape. It's certainly also simple with a perfect equilateral triangle, or some other well-defined shapes.

But, if you are using any kind of shape (at least 3 points), the problem is harder. You need to know what exact proportions you have to find on the result, the one that provides you with the exact original not-distorted shape. Even if the shape is a rectangle, it's already hard to be sure to restore its exact original height/width ratio. I do not see an other comfortable way to do this than having this reference shape as a third image (a square, a rectangle, a triangle, whatever, etc).

Perhaps the best way for you to understand all this is to practice a bit with real cases...

;-)


Antonio F.G.
 

On Wed, Apr 28, 2021 at 12:55 AM, Etienne Monneret (Perso) wrote:
Can't figure out how to understand your drawing. At least I need some
explanations.
Above is a real chequerboard (without pieces). In the middle is a pair of stereo cameras represented by their focal points and their sensors. An image of the chequerboard is captured in the sensors. Below is the phantogram plane, on which the images of the sensors are re-projected.

If the baseline is parallel to the chequerboard plane, and the phantogram plane is also parallel to that plane, then the two re-projected images are IDENTICAL, which is what we are looking for, is not it?


Not sure to understand you. Whatever the shooting angle, the final
viewing will be at the same angle.
Yes, this was what I meant


But, if you are using any kind of shape (at least 3 points), the problem
is harder. You need to know what exact proportions you have to find on
the result, the one that provides you with the exact original
not-distorted shape. Even if the shape is a rectangle, it's already hard
to be sure to restore its exact original height/width ratio. I do not
see an other comfortable way to do this than having this reference shape
as a third image (a square, a rectangle, a triangle, whatever, etc).
I see what you mean. But this looks a very cumbersome procedure, impossible most of the times.
You can perhaps get a reference image for small scenes in your table top, but what if you are shooting in a street or other big place? How do you get the reference image? Drones? Surveying with a theodolite?

I am not sure whether this will work, but my bet is to find the projection that nulls the vertical and horizontal disparities of a set of matching points on the the ground scene. If the images are identical the disparities are obviously zero, and my hope is that a re-projection with zero H & V disparities needs to have identical images.
The matching points need not follow any shape, rectangle, triangles et al. they only need to belong to the same plane. This is an small step further to the alignment algorithm of StMani3 that finds a re-projection that nulls only the vertical disparities.
I am adding a patch to StMani3 to test this. Stay tuned:-)

Regards
     Antonio


Antonio F.G.
 

On Wed, Apr 28, 2021 at 10:20 AM, Antonio F.G. wrote:
I am not sure whether this will work, but my bet is to find the projection that nulls the vertical and horizontal disparities of a set of matching points on the the ground scene. If the images are identical the disparities are obviously zero, and my hope is that a re-projection with zero H & V disparities needs to have identical images.
I am adding a patch to StMani3 to test this. Stay tuned:-)
Well, I have already introduced in StMani3 the goal to null the vertical and horizontal disparities at the same time.

The bad news is the successive approximation process gets lost and crashes.
The good news is that https://stereopix.net/photo:9ON3DvIwow/ still proves that this IS POSSIBLE. Only have to try another way.

Regards
    Antonio