[x264-devel] Question on the motion vector resultion

Erich Marth destinationcloud at googlemail.com
Mon Sep 7 18:51:27 CEST 2009


So the Value MVP presents qpel's. So the fullpel Position 1 is Four  
(One fullpel x 4 quarters per pel)

I Know that the refinment is After the normal estimation. But my  
question was how the MVP on qpel's Position is Done in the First step


Erich

On 07.09.2009, at 18:27, Jason Garrett-Glaser <darkshikari at gmail.com>  
wrote:

> On Mon, Sep 7, 2009 at 8:45 AM, Erich Marth
> <destinationcloud at googlemail.com> wrote:
>> Hi there
>>
>> I do have a question. I'm doing some research on the motion  
>> estimation
>> algorithm.
>> For each slice, every single macroblock is analyzed.
>> Based on the SLICE_TYPE, the analyzing is done on intra or inter  
>> decisions.
>>
>> In the P/B slice case
>> we do first calc the cost for all valid mv cases.
>> in the x264_macroblock_analyze method we try to find the best match.
>> But first we are going to predicted the needed motion vector from  
>> available
>> neighbours.
>> In H.264 the motion vector's resolution is able to do quarter pel  
>> positions.
>> But the mvp which is calculated in the mv fetch  
>> (x264_mb_predict_mv_16x16)
>> method is of type integer?
>
> It isn't; it's qpel.
>
>> Why? I mean if all neighboring macroblocks do have quarter pel  
>> positions
>> the current mvp shall be on a quarter pel position too? So why  
>> integer?
>
> It isn't.
>
>> And why didn't I find any half/quarter pel estimation in diamond  
>> search ...
>> ?
>
> Because subpel motion estimation is in refine_subpel, not
> x264_me_search_ref, hence the name.
>
> Dark Shikari
> _______________________________________________
> x264-devel mailing list
> x264-devel at videolan.org
> http://mailman.videolan.org/listinfo/x264-devel


More information about the x264-devel mailing list