Re: What is currentMV and predMV in SearchData struct

There is no "matching Macroblock in previous Frame". The vector found by
ME can point to arbitrary pixel (and sub-pixel) positions in the reference
frame and not just to macroblock borders.

Therefore, there isn't a single matching macroblock. The reference samples
used for motion compensation may span over several macroblocks. You could
just determine the macroblock which most closely matches a given pixel
position by rounding to nearest macroblock border (and also clipping into
the [0, mb_width] and [0, mb_height] area). But I don't see for what you
would need to do this?

Regards,
Michael


Quoting Chien Tran <[email protected]>:

> Thanks for your answer.
> So let's say I found the match_x and match_y of referenced Macroblock. How
> could I get the exact Macroblock from the macroblock pointers (pMBs)?
> Currently I see some places in the code do like this:
> pMB = pMBs[x + y * mb_width]; //get Macroblock based on x, y location
>
> But I am afraid I can not do like that if (x + y * mb_width) < 0, right?
>
> 2012/5/21 Michael Militzer <[email protected]>
>
>> Hi,
>>
>> your understanding is correct: currentMV stores the best candidate vector
>> found so far during ME for a respective mode/partition (e.g. 16x16) and
>> predMV holds the corresponding predictor vector that will be substracted
>> before actual coding of the vector.
>>
>> Note that only the difference between best found vector and predictor
>> vector is written to the bitstream (refer to e.g. get_pmv2() in
>> mbprediction.c on how the predictor vector gets determined). Considering
>> the predMV already during ME is important to estimate the motion vector
>> cost (so the amount of bits needed to write the found vector into the
>> bitstream). The more currentMV deviates from predMV, the higher the cost
>> for coding the vector into the bitstream.
>>
>> Also note that motion vectors can indeed point outside the visible
>> rectangle of a ref frame. So your understanding on how to determine the
>> reference blocks for motion compensation is correct but you need to
>> consider that there is a padded, "edge" area around reference frames.
>> Therefore, "match_x < 0" is legal. You can take a closer look at how
>> the edge padding is performed in image_setedges() (image/image.c).
>>
>> Best regards,
>> Michael
>>
>>
>> Quoting Chien Tran <[email protected]>:
>>
>> > Hi everyone,
>> > I am new to Xvid and I am doing a research about video compression. When
>> I
>> > look at the SearchData struct, I saw two important fields: currentMV and
>> > predMV. I guessed it stored the information about the motion vector of
>> > Macroblock, but I don't know what it really means to have 2 fields like
>> > that. Its comment said currentMV is "best vectors found so far"
>> > (currentMV[0] would be the MV for the 16x16 MB, I think) and the predMV
>> is
>> > "vector which predicts current vector". Could you guys please tell me
>> what
>> > currentMV and predMV are really used for?
>> > Let's say I have the x, y of the current Macroblock. How can I know the
>> > location of the matching Macroblock in previous Frame that were found in
>> > Motion Estimation? Is it: match_x = x + currentMV[0].x; match_y = y +
>> > currentMV[0].y? I did some tests and found that it sometimes has the
>> value
>> > that is out of the Macroblock array (more specifically, match_x < 0).
>> >
>> > --
>> > Best wishes,
>> >
>> > Tran Xuan Chien
>> > University of Information Technology
>> > Phone: (+84) 1692 468 154
>> > _______________________________________________
>> > Xvid-devel mailing list
>> > [email protected]
>> > http://list.xvid.org/mailman/listinfo/xvid-devel
>> >
>> >
>>
>>
>>
>>
>> _______________________________________________
>> Xvid-devel mailing list
>> [email protected]
>> http://list.xvid.org/mailman/listinfo/xvid-devel
>>
>
>
>
> --
> Best wishes,
>
> Tran Xuan Chien
> University of Information Technology
> Phone: (+84) 1692 468 154
> _______________________________________________
> Xvid-devel mailing list
> [email protected]
> http://list.xvid.org/mailman/listinfo/xvid-devel
>
>