msm_007
18th February 2009, 03:03
Hi all, I already sent this message in a sub subject, but I think it merits a new subject
My aim is to implement a parallel application of video coding who conserve the same performance in size of result file (raw byte stream in my case .264) but more faster than other applications. I’m not going to implement a new version of coders or modify another applications, I’m just going to use one of those applications (the famous x264, I think it is the best) in parallel paradigm
To do that I decompose the YUV sequence in a number of small sequences and then every sequence it going to be coded separately on a different machine, then I’m going to join the byte stream of every coded sequence to have the finale .264 file. For example, decompose a file of 300 frames on 5 files of 60 frames.
It worked but the problem is that the size of the result after join is 3 times larger than the result if I proceed with the 300 frames without decomposition
I guess the problem is on the border frames (1 files gives 2 borders but 5 files gives 10 borders witch degrades the compression performance)
The only solution I have in mind to deal with the size problem is to code the borders with bframe: send redundant frames to use them on coding the borders
E.g. for the 2st file, 60-119,
- Send the 5 frames (55-59) with the portion and start coding from the frame 60 and use them to code the 1st 5 frames (60-64) and the same send (120-124) to code the last 5 frames (115-119)
==>The result is 60 frames (start = 60, end =119) coded and saved in the file, all of them are coded with bframe and ref=5
But I don’t know if it is possible to implement such idea
Any ideas Pleeeeeeeeeeeeeeease?
My aim is to implement a parallel application of video coding who conserve the same performance in size of result file (raw byte stream in my case .264) but more faster than other applications. I’m not going to implement a new version of coders or modify another applications, I’m just going to use one of those applications (the famous x264, I think it is the best) in parallel paradigm
To do that I decompose the YUV sequence in a number of small sequences and then every sequence it going to be coded separately on a different machine, then I’m going to join the byte stream of every coded sequence to have the finale .264 file. For example, decompose a file of 300 frames on 5 files of 60 frames.
It worked but the problem is that the size of the result after join is 3 times larger than the result if I proceed with the 300 frames without decomposition
I guess the problem is on the border frames (1 files gives 2 borders but 5 files gives 10 borders witch degrades the compression performance)
The only solution I have in mind to deal with the size problem is to code the borders with bframe: send redundant frames to use them on coding the borders
E.g. for the 2st file, 60-119,
- Send the 5 frames (55-59) with the portion and start coding from the frame 60 and use them to code the 1st 5 frames (60-64) and the same send (120-124) to code the last 5 frames (115-119)
==>The result is 60 frames (start = 60, end =119) coded and saved in the file, all of them are coded with bframe and ref=5
But I don’t know if it is possible to implement such idea
Any ideas Pleeeeeeeeeeeeeeease?