[x264-devel] {Spam?} Re: Multislicing support

[Etienne Bömcke]

It seems like I wasn't as clear as I thought. Correct me if I'm wrong,  
but I think that the H.264 standard only forbids inter-slices  
prediction inside of one frame. Removing a slice of one frame ony  
shouldn't cause any harm to the decoding process of the rest of the  
frame.
The option I added (no-mv-inter-frames, complicated name but it was  
the best I could come up with. If anybody has a better idea, I'm open  
for suggestions) allows one to encode a sequence in two or more  
completely independant slices. That means that you could remove a  
given slice from the whole sequence while the remaining slices could  
still be decoded.
Basically I modified the motion estimation related code and restrained  
the search range so that a macroblock of a given slice can only be  
predicted from a macroblock in the same slice, even when predicting  
from previous frames.
As I said in one of my first e-mails, the objective for this task is  
to be able to replace a part of an H.264 sequence by a whole different  
sequence directly in the bitstream. This restriction allows one to do  
that without any decompression error cause by the slice replacement.

I hope I'm clear enough this time, if not do not hesitate to ask me  
further explanations.

Regards,

Etienne

PS : Thanks for your comments, I really appreciate it.




On 19 Feb 2008, at 10:05, List, Peter wrote:

>
> Dear Etienne,
> Thank you very much for taking this task! I believe this was an  
> important effort for x264! (Even more so, if you will manage to  
> split a frame into slices of specific byte sizes).
>
> Just one comment to your "--no-mv-inter-slices" option.
> To my knowledge the H.264 standard says, that inter-slice prediction  
> is forbidden in ALL CASES! To my understanding it is the very  
> meaning of slices to...
>
>> allowing one slice to be completely removed without causing any  
>> harm to
>> the decompression of the remaining ones.
>
> ...as you said below. There is no other reason for having slices at  
> all.
> Most of all, that would mean, if one would turn this option off, the  
> resulting bitstream would NOT BE H.264 complient!
>
> Probably I missed something! Please explain.
>
> Best regards
> Peter
>
>
>> -----Ursprüngliche Nachricht-----
>> Von: Etienne Bömcke [mailto:etienne.bomcke at uclouvain.be]
>> Gesendet: Montag, 18. Februar 2008 15:11
>> An: Mailing list for x264 developers
>> Betreff: Re: [x264-devel] Multislicing support
>>
>> Ok, so here's the patch I created to add support for multiple slices
>> in one frame. I use the following two command-line options : --mb-
>> slicesize and --no-mv-inter-slices. The first option allows one to
>> specify the desired slice size in number of macroblocks, while the
>> latter forbids any inter-slice prediction, allowing one slice to be
>> completely removed without causing any harm to the decompression of
>> the remaining ones. I tested it the way Lorren described in his e-
>> mail. Comparing x264's fdec.yuv file and the output of the standard
>> decoder (actual version, 13.2 I think) using the 'diff' utility gives
>> a blank result, which means that both files are identical.
>> However, I haven't been able to test Mojtaba's part, ie the splitting
>> of frames in multiple NALs when specifying a maximum NAL size in
>> bytes. If anybody could take a look and tell me if it seems clean,  
>> I'd
>> appreciate it.
>>
>> One last thing : You'll notice that I changed the macroblock scanning
>> condition in file encoder.c . Seems to me that the last macroblock of
>> the slice (stored in sh->i_last_mb) should be the last macroblock
>> contained in the slice, and not the first one just outside. I checked
>> for further reference to that variable, but it seems like it's not
>> used a lot, so this modification didn't break anything.
>>
>> Etienne
>>
>
> _______________________________________________
> x264-devel mailing list
> x264-devel at videolan.org
> http://mailman.videolan.org/listinfo/x264-devel

--
Etienne Bömcke
UCL - Laboratoire TELE
etienne.bomcke at uclouvain.be