Jump to: navigation, search

Autopano Video - Stereo (3D)



Presentation


To see the relief our brain uses offset visual spots of our 2 eyes. To recreate the depth perception on a video, you will have to show two different images, one for each eye. Then our brain will understand this offset called parallax, as if it was deepness.



Disparity = 3D


The different point of view between our 2 eyes is understood by a different position of objets on the left image and on the right image (see image below).
This shift (change) of positions is called disparity. And more specifically the horizontal shifts give you the perception of 3D topography.
The disparity is materialized by a cyan and red fringe all around the objects. The more the disparity is large, the more the 3D feeling is intense.

 Caution, above a certain range of 3D topography the viewer will be disturbed. You must stay reasonable about the amount of 3D topography


Disparity

In the example that you can see below, the buildings in the background don't have any disparity. This is totally normal because they are far away. The person in the front has an important disparity which means that it will not be blended into the background. That's the effect we are looking for. We must remind you that all the objets at equal distance from the rig have the same disparity.



Staging of 3D


The director can use two settings to set. The first one is the distance between 2 points of view that affects the amount of 3D topography and the second is the convergence that changes the deepness of the scene.

In the particular VR case you must set the center-to-center distance and the convergence as follows : « center-to-center distance at 65mmm » and « convergence at infinity »
Unfortunately, these settings can't be automatically determinded in AVP. Nevertheless AVP allows you to ajust these 2 settings with the center-to-center distance (Interocular) slider and the value of convergence. Please refer to the section 3D settings in AVP for more details.
When it comes to broadcasting a 360 video on a screen the problem is more complex and we won't give explanations about this subject.




3D settings in AVP


Stereo menu:
Stereo menu

Be sure you have ticked the Stereo mode box to activate this mode



Assigning cameras


This setting will let you define the strategy which selects the best images for the left eye rendering, then the right eye (see picture below)

Camera assignement


This strategy is chosen independently for each camera, so it is possible to manage various RIG configurations. To define this strategy, just click on the Stereo icon in the toolbar to display the Stereo settings.

As shown in the Stereo menu image above, you can see the list of cameras. For each of these, you can choose one of the 7 strategies for the final render:

  • Assignment:
    • Left and right : this camera will be used as much as possible for the two renders (very useful for zenith and nadir)
    • Left only : this camera will be used as much as possible for the left render. If you do have a choice, prefer one of the Split modes (see below)
    • Right only : this camera will be used as much as possible for the right render. If you do have a choice, prefer one of the Split modes (see below)
    • Vertical split : the right part (resp. left) of the image will be used for the left eye (resp. right). The delimitation depends on the Interocular setting. We recommend to use this setting for circular rigs with cameras placed in landscape mode.
    • Vertical split inverted : the left part (resp. right) of the image will be used for the left eye (resp. right). The delimitation depends on the Interocular setting. We recommend to use this setting for circular rigs with cameras placed in landscape mode, upside down.
    • Horizontal split: the upper part (resp. lower) of the image will be used for the left eye (resp. right). The delimitation depends on the Interocular setting. We recommend to use this setting for circular rigs with cameras placed in portrait mode.
    • Horizontal split inverted: the lower part (resp. upper) of the image will be used for the left eye (resp. right). The delimitation depends on the Interocular setting. We recommend to use this setting for circular rigs with cameras placed in portrait mode.




Center-to-center distance (Interocular)


 Caution: the center-to-center distance setting applies only for circular cameras in Split mode. The center-to-center distance is the distance between two
points of view.



I would point out that to create a proper and realistic 3D topography with a HMD, the center-to-center distance must be as closely as possible to the human eyes, usually 65mm. However, AVP propose a center-to-center distance setting without standards of measurement (not in mm). Indeed, the calculation of the center-to center distance in mm depends on the RIG configuration which would require a complex calibration.

The center-to-center distance setting between:

  • 0%: center-to-center distance is zero, no 3D
  • 100%: center-to-center distance as it maximum depending on the available overlapping


Entraxe

Increasing center-to-center distance increase 3D topography.

The picture above shows the difference between a center-to-center distance set to 50% and 100%. You can see the disparity (cyan and red fringe ) visible on the character is changing, whereas it is not changing on the buildings in the background.

By default, AVP sets the center-to-center distance at 100% in order to:

  • 1 – maximize 3D topography effect and
  • 2 – avoid areas without any 3D effect


In practice, this value set at 100% gives the best results with some RIGs we tested. However, if this default value does not suit you, you can reduce center-to-center distance using the Interocular  slider.



Convergence


 Caution: the entraxe setting applies only for circular cameras in Split mode.


Just below the Interocular  (center-to-center distance) slider, AVP let the possibility to correct the convergence angle (in degrees). This parameter is modifying the reference convergence angle in order to change the convergence distance.
It is important to bear in mind that objects at a convergence distance will have zero disparity (no colored fringe).

In the case you make a production for a HMD, you must set the convergence to infinite. With this configuration the optical axes are parallel and the convergence angle is equal to zero. This means you must configure the convergence so that the infinitely distant objects appear without disparity within the image (no colored fringe)
The example below shows the convergence is incorrectly set as you can see a great disparity in the backgroung (buildings)

Convergence

Changing the converge moves the scene back and forth.

AVP will let you change the convergence to correct it if needed. This corresponds to a deviation in degrees of the convergence angle against a reference convergence. You must change this value until the disparity on distant objects disappear (colored fringe).
In the example below, the convergence angle was choosen to remove the disparity on the buildings.

Convergence

Changing the converge moves the scene back and forth.
By default, we let the convergence to 0° to have the convergence visible on the different areas we use to stitch the pictures.



Rendering modes


At the bottom of the Stereo window, you can choose which stereo setting you want to apply. This choice is related to the diffusion method:

  • Left-Only : only render the left eye
  • Right-Only: only render the right eye
  • Over/Under: render both eyes displayed below each other. This is the most popular mode in 360-3D (Youtube, …)
  • Side by Side: render both eyes displayed next to one another
  • Anaglyph: this is useful to manage the relief in the preview window but not appropriate with a HMD



Click Apply to save the settings:

  • Camera assignation
  • Center-to-center distance value
  • Convergence value
  • Rendering mode

Once validated, it is applied to the preview window.



Advices


To be sure you will get the best results, we recommend to use a RIG with 5 or 7 cameras placed in a circular manner (i.e. GoPro Odyssey, Google Jump). You can add camera to the zenith or nadir where the stereo vision is not the important part.

For best results in relief, we recommend to use the Split mode (refer to Assigning cameras above). Indeed, this means a better 3D stereo vision because the most appropriate camera will be used. However, using this mode will create more cutting zones where the parallax can be a problem (especially on close objects). As it is often the case, it requires some compromises. Consider whether the 3D vision quality ot whether the stitching quality.




References


For more details about stereo shooting techniques, please refer to 3D Movie Making: Stereoscopic Digital Cinema from Script to Screen







BACK TO: Documentation / Autopano Video Documentation