Complexity of the encoder

[koliva]

Hi experts,

I am trying to compare the complexities of different recent state-of-art video encoders. For that purpose, I need some information about complexity distribution over different processes such as motion estimation, motion compenstaion, quantization, entropy coding.

It would be the best if you can give me some reference articles where the complexity percentages of those processes are investigated, i.e. %70 goes for motion estimation, %10 for motion compensation, %10 for qantization and %10 for entropy coding (I just made up those values).

Thanks.

[MatLz]

Instead of a cross-thread, why not a small bump here ? :
http://forum.doom9.org/showthread.php?t=155552

[koliva]

Instead of a cross-thread, why not a small bump here ? :
http://forum.doom9.org/showthread.php?t=155552

My post waited there 20 days and I got no reply. Then I thought that not all experts are looking at newbies topic. This is the reason why I post it also here.
Any answers related to my question?

[Reimar]

Any answers related to my question?

I'd say: any profiling tool should be able to answer your question for a specific case.
It's not really possible to answer in general in a truly useful way, at least the number of CPUs available must be known, and whether the "wall clock" time or cpu-time is relevant, what kind of parallelization the encoder uses, CABAC vs CAVLC...
In principle, the more processing units you throw at it, the more the bitstream coding becomes an issue since it's not really possible to do more than frame-level parallelization for that - unless you're willing to pay the cost in compression efficiency to use lots of slices for H.264.
I admit I couldn't give you an answer anyway, but if you gave the _purpose_ of the question, maybe someone could at least figure out the question you really want answered, because the one you gave is only really answerable with basically made-up numbers.

[J_Darnley]

koliva, perhaps you want to look at some of the profiling done here: http://akuvian.org/src/x264/

[Shevach]

There are few articles where computational complexities (or performances) of different stages of H.264 encoding are analyzed. Most of articles are dedicated to coding efficiency.

In the paper "ANALYSIS AND DESIGN OF MACROBLOCK PIPELINING FOR H.264/AVC VLSI ARCHITECTURE" by Tung-Chien Chen, Yu-Wen Huang, and Liang-Gee Chen, you can find a diagram where performances of all encoding phases are present.

More article related to performances is:
"Real Time H.264 Video Encoder Implementation on A
Programmable DSP Processor for Videophone Applications", by Shashi Kant, Mithun U and PSSBK Gupta.