H.264 MVC (3D) Encoding efficiency - question

Frantisek_K_ · ‎01-21-2014

Hello,
I noticed strange phenomenon:

When encoding MVC-3D video using Intel Media SDK (e.g. sample_encode) - full HD, but it is probably not related,
and using the same YUV input for both L-eye and R-eye,
then I would expect that resulting dependent view h264-stream (in viewoutput mode) will be very small, because difference between eyes is zero.

But it is not truth - dependent view h264-stream is quite big, almost the same size and base-view!

Is that correct?
Can be this ratio base:dependent view somehow influenced?

Thanks

Frank

Frantisek_K_ · ‎01-22-2014

Just to add more info: I'm working with Intel Media SDK 2013 R2, and encoder is set to variable_bitrate mode.

Anthony_P_Intel · ‎01-22-2014

Hi Frank,

We will look into this (VBR encoded to MVC) case for coding efficiency benefits of future products. Thank you for your input.

Frantisek_K_ · ‎01-23-2014

Thanks - looking for future evolution of IM SDK :)

Just to give you additonal info about current SDK results (example):

Input: Left eye - YUV full HD (1920x1080, 23.976 fps), 275 frames
Right eye = identical to left eye

Encoder:
Intel Media SDK 2013 R2 - software libmfxsw32.dll
(but Intel Media SDK 2014 for Clients gives the same output)

- h.264 - MVC 3D
- variable bitrate - target 17 mbit/s, maximum 28 mbit/s
- viewoutput

Output:   base_view.h264          13.865.861 bytes
          dependent_view.h264     10.086.666 bytes

Since left and right eye are identical, there is no logical reason for dependent_view.h264 to be so big !
Typically 3D Bluray's have ratio base:dependent about 2:1 for regular 3D movie.

Cheers,

Frank