How to configure JM software to encode multiview images

[koliva]

Hello experts,

We are currently trying to encode our video streams captured by a camera array in which the cameras are located horizontally and vertically. For that purpose, we are looking for a scheme to apply on it. Is there any (traditional) way to encode this huge amount of data efficiently?

There is a scheme in our mind but it seems to us that it is not possible to configure JM software according to that scheme.

Scheme:

http://img28.imageshack.us/img28/9773/scheme1.png

Any help will be appreciated.

[Guest]

You would have to write your own encoder/decoder to support a scheme like that. I know of no existing encoders that support anything other than a single linear chain of frame dependencies.

[koliva]

You would have to write your own encoder/decoder to support a scheme like that. I know of no existing encoders that support anything other than a single linear chain of frame dependencies.

I can rephrase this scheme as a single chain:

http://www.yukleresim.com/images/drt2wsjj9vln8z8ugic.png

Still not possible? Even in JMVC?

[Guest]

You're kidding yourself. Moving things around does not change the topology.

[Dark Shikari]

He means an MVC stream with 5 views per frame. That doesn't seem particularly unreasonable and can probably be done with the JSVM reference encoder.

[chen]

Will someone please give me some direction on how to configure the JM reference software for eight views? What is the default prediction structure? I found no description of the MVC mode in the documentation, though it is said there is.