Re: [Dev] [3dweb] Multi-scopic displays


Here are a few very quick observations:

1) Auto-steroscopic displays tend to require more than 2 pictures. Alioscopy for example requires 8, Tridelity usually 5. This is not true for mobile devices, where two pictures are usually sufficient because there is a single viewer and you are not trying to build parallax.

2) There are two big philosophies on how to generate the many pictures.

2a) The "pure" approach of Alioscopy or Tridelity is that you can send completely independent pictures.

2b) The "generated" approach of Philips/Dimenco, Dolby or Toshiba is that you extrapolate the pictures from either a depth map and color map or a stereoscopic pair

2c) The benefits of the pure approach is that it has no problem with transparency or parallax. It never needs to "invent" a pixel that is not there, whereas the "generated" approach does. With a depth map, you can only represent one depth, so no transparency either.

2d) The benefits of the "generated" approach is that it works better with existing infrastructure. By contrast, a mixed Alioscopy frame needs to be pixel-perfect for a good 3D effect, and are built in such a way that compression won't work well, see for an example.

[Taodyne's approach is to have a device-independent representation, and render for a specific device at the last minute. For example, we may have 8 different images and mix them for Alioscopy, or take 5 out of 8 and mix them for Tridelity, or take 2 out of 8 and interlace them for a passive TV, or 2 out of 8 and frame-sequence them for an active TV. This lets us leverage existing encodings and transports.]

Now specifically about the proposal at, and ignoring multiscopic displays:

0) Does this apply to a contents or does it specify how it is rendered? For an image, you sometimes need to know its format, i.e. it's a JPG with side-by-side pictures. But for a rendered element, e.g. a text or some 3D object, you need is to describe how it will be rendered. Converting existing assets may be technically hard & expensive, but not always. So it makes sense to say "I have a side-by-side stereo image and I want to show it on an interlaced display", but not "I have a stereo-picture with a baseline of X and I want to show it with a baseline of Y"... unless this is a synthetic 3D picture (e.g. a .obj file)

1) perspective-baseline: The issue raised for perspective-baseline cannot really be answered without a camera / frustum model for what we are looking at. In addition to the baseline, we also need to know how the virtual eyes converge to some point (or to infinity).

2) stereo-contents: What happens if you have a format that already specifies that? For JPG, you have JPS (2 pictures) and MPO (n pictures). I understand what a JPG with side-by-side stereo means, but what does a JPS with side-by-side stereo mean? How would I put a JPS or an MPO in a web page?

3) stereo-render-option: Besides being hard to implement, I personally have no clue what "center" would mean precisely, unless there's already a center image (e.g. with a 5-pictures multiscopic format).

4) stereo-format: additional output formats include frame-sequential, checkerboard and anaglyph. For input, there are also formats that specify multiple pictures themselves. We could probably call that frame-sequential as well.

Just my two cents for today
Best regards

On 26 févr. 2013, at 05:37, "Dong-Young Lee" <> wrote:

> Hi, Christophe,
> The spec proposal was written specifically "stereoscopic" 3D in mind, since
> stereoscopic displays became quite popular these days (e.g., 3DTV).  I
> haven't thought much about autostereoscopy or multiscopy, but there should
> be things to be considered for multiscopy at least.  I'm definitely
> interested in those topics and would welcome your input and participation.
> Best Regards,
> Dong-Young Lee
> -----Original Message-----
> From: Christophe De Dinechin [] 
> Sent: Tuesday, February 26, 2013 3:18 AM
> To:
> Cc:
> Subject: [3dweb] Multi-scopic displays
> I've quickly browsed through the 3dweb proposal at
> I wanted to
> know if any consideration had been given to auto-stereoscopic and
> multiscopic displays (e.g. Alioscopy, Dimenco or Tridelity)? I can't find
> anything by searching through the archive.
> Taodyne supports this kind of displays and the description of 3D scenes
> (including mixing 3D movies and real-time 3D object rendering). Here is an
> example:
> nts. 3D documents can be rendered on 2D or 3D displays, including
> glasses-free displays. We can share our experience on this list if there is
> any interest.
> Best regards
> Christophe de Dinechin
> _______________________________________________
> Dev mailing list

Received on Monday, 4 March 2013 07:45:14 UTC