[x264-devel] Re: Rate control in x264

[Son Minh Tran]

Thank you for your precision. I just want to have your confirmation once 
again
So in order to have the x264 encoder (as a whole encoding chain) 
independent from the entropy coder, I have two choices to run x264 as 
followings:
1) ./x264 -b 4 -r 7 --mixed-refs -A all -8 -qp [0..32] -o <Output> 
<Input.yuv> <width>x<length>
2) ./x264 -b 4 -r 7 --mixed-refs -A all -8 -crf [0.0 ... 1.0] -o 
<Output> <Input.yuv> <width>x<length>
I use also -r 7 (max reference frameNo) --mixed-ref (block 8x8 can use 
other RefFrame than the neighbour block in the same MB) and -A all -8 
(chack all possible types of block) to maximize the advanced features of 
h264.
My question is still with the addition of  -qp N or -crf  R  will I have 
the independence between the entropy coder and the function units before 
it. By defaut, qp is set to 26. Is it mean that  I already have this 
independence without explicitly specifying it at the command line.
The trellis setting (-t N) is disabled by defaut. How can I check wether 
the rate distortion optimation is off? I think here you mentioned the 
proper value of the option -m (with 7, x264 deploys the real RDO ). By 
defaut it takes value 5, is it small enough?
Thank you
  
Loren Merritt a écrit :
> On Mon, 21 May 2007, Son Minh Tran wrote:
>
>> I try some tests on the performance of CABAC. Then I need exactly the 
>> same input data for CABAC encoder. Because of the rate control of 
>> x264, I do think that I don't have the same quantized data (before 
>> CABAC encoding phrase) of the frame T if in the previous frames, the 
>> output of the whole coder
>> (after CABAC encoding phrase) was different.
>> I know that there are 5 models of rate control in x264. What I need 
>> is the constant quantizer. Would you please show me how I can 
>> activate this mode? Is it enough to use --qp N at the command line?
>
> Yes. And crf mode also works in that it doesn't take any feedback from
> the actual size of the encoded pictures, so its choices will be
> independent of the entropy coder.
> But ratecontrol isn't the only way the entropy coder can influence
> encoding decisions. Also make sure not to enable rate-distortion
> optimization or trellis.
>
> --Loren Merritt
>

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html