[Mp4-tech] [video] Error concealment - unreferenced memory error

[Steve]

Hi Adarsh,
Thank you for your quick and very helpful reply, to know that I was heading
in the correct direction is great news!
I moved my code to decode_one_slice as suggested (I inserted it after
read_flag = read_one_macroblock(img,inp); and before
decode_one_macroblock(img,inp);), unfortunately I was still getting the
same error on the same macroblock of the same frame.
I had IntraPeriod=0 set in encoder.cfg, frame 18 was a P frame. As your code
only worked on I-Frames I incremented the number of I-Frames in the encoded
file and even with just one additional I-Frame at frame 10 the error
disappeared.
Personally I would not have thought it unreasonable to have 18+ frames
between I frames, incrementing the number is acceptable for my work for the
time being however does not really solve why the error was occurring in the
first place.
I will keep looking into this problem, however if your good self or any
other experts can shed any further light on the problem then it would be
very much appreciated.
Thanks again for your help,
Steve Beesley
Quoting Adarsh Golikeri <golikeri gmail.com>:
> hi steve,
>
> i've tried this same method before. the concept you are using is
> correct (setting the ei flag to 1) but the place you are doing it is
> wrong. instead of in function exit_picture, try doing it in the
> function decode_one_slice. this was the code i used :
>
>     // lose some MBs in i-frames
>     loss = rand()%100;
> 	if (img->number>0 && img->type == 2 && img->current_mb_nr >= 11 &&
> img->current_mb_nr <=22)
>        {
> 	   img->mb_data[img->current_mb_nr].ei_flag = 1;
>        fprintf(stdout,"\nLosing MB %d of Picture
> %d",img->current_mb_nr,img->number);
> 	   }
>     // lose some MBs in i-frames
>
> hope that helps,
>
> adarsh
>
>
> On Mon,  7 Feb 2024 16:09:30 +0000, Steve <S.T.C.Beesley lboro.ac.uk>
> wrote:
> > Hi experts,
> >
> > I have been working with JM93 to look at error concealment,
> specifically the
> > way that the decoder tries to hide any missing data. In order to
> achieve
> > this I have been adding something similar to:
> >   img->mb_data[207].ei_flag = 1;
> > in image.c after line 1206:
> >   ...
> >   //! mark the start of the first segment
> >   if (!dec_picture->MbaffFrameFlag)
> >   {
> >     ercStartSegment(0, ercSegment, 0 , erc_errorVar);  // <-- 1206
> >   ...
> >
> > For the most part this method seems to work extremely well however
> under
> > some circumstances the decoder is throwing an unreferenced memory error
> due
> > to the Motion Vector reference slice being set to -1. One such example
> of
> > this is in a cif version of mobile.yuv when ProfileIDC = 100 (High) is
> used
> > at macroblock 207 (w=9, h=9), frame 18 (all frames up to 15 in the
> decode
> > sequence work ok).
> >
> > >From what I can tell there are only two lines that ever set the mv[2]
> > variable to anything other than 0 and these are again in image.c:
> >   pRegion->mv[2] = (dec_picture->ref_idx[idx][ii][jj]);
> > or
> >   pRegion->mv[2] = dec_picture->ref_idx[LIST_0][ii][jj];
> > depending on the type of slice (non-B or B).
> >
> > I'm wondering if anybody can shed any light onto why this variable
> would
> > ever be set to -1, judging by the rest of the EC code it looks like it
> > assumes that this variable is always going to contain a valid number as
> no
> > checking appears to be done. One working theory I currently have is
> that it
> > could be referencing a slice that has either not yet been decoded or is
> no
> > longer in the decode buffer. Also can anybody suggest where to find the
> > information to detect if the returned value is valid or not and what
> action
> > should be taken in the cases where it is invalid?
> >
> > If it is the case that I'm barking up the wrong tree here (which is
> more
> > than possible) then can anyone suggest a better method of forcefully
> adding
> > errors at the decoder?
> >
> > Many thanks in advance,
> >
> > Steve Beesley
> > _______________________________________________
> > NOTE: Please use clear subject lines for your posts. Include [audio,
> [video], [systems], [general] or another apppropriate identifier to
> indicate the type of question you have.
> >
> > Note: Conduct on the mailing list is subject to the Antitrust
> guidelines found at
> http://www.mpegif.org/public/documents/vault/mp-out-30042-Antitrust.php
> >
>