[FFmpeg-devel] [PATCH 0/3] add vvc raw demuxer

[James Almer]

On 12/15/2020 10:43 AM, Nuo Mi wrote:
> On Tue, Dec 15, 2020 at 4:44 AM Mark Thompson <sw at jkqxz.net> wrote:
> 
>> On 14/12/2020 13:31, Nuo Mi wrote:
>>> Hi Mark,
>>> I have almost done the cbs for sps, pps, and slice header. I will start
>> to
>>> implement the parser.
>>
>> This looks fun :)
>>
>>> Few questions for you:
>>> 1. We need over-read some nals to detect the frame boundaries. But those
>>> nals may switch/replace sps/pps. Do we need to use an output cbs and get
>>> frames from the output cbs?
>>
>> I'm not seeing where this can happen - if you see a new parameter set then
>> you must be in a new AU so you don't parse it, while a new VCL NAL as part
>> of a new frame with no PS before it must have a parsable header?  (Or am I
>> missing some problematic case?)
>>
> 
>>> 2. We can't handle an incompleted nal in current cbs.
>>>
>> https://github.com/FFmpeg/FFmpeg/blob/03c8fe49ea3f2a2444607e541dff15a1ccd7f0c2/libavcodec/h2645_parse.c#L437
>> ,
>>> do we have a plan to fix it? What's your suggstion for the frame split?
>>
>> I'm unsure what the question is.  You will always need to keep reading
>> until you split a valid NAL unit which isn't in the current AU (noting that
>> a slice is the last slice in the current frame is never sufficient, because
>> suffixes might follow).
>>
>> I want to feed the input buffer to cbs and get nal units from it. Seems it
> does not work like this.
> We still need a frame split in the parser. And feed the entire frame to
> cbs.

The parser is meant to look byte by byte until it has found enough data 
to complete an entire Access Unit worth of NALUs.
See how hevc does it in 
https://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavcodec/hevc_parser.c;h=463d352055b971bdcc27a27b3fcd7aa4b609286e;hb=HEAD#l261 


Once the delimitation of an AU is found, then the resulting buffer can 
be parsed (In VVC's case, we want that to be done by CBS) and ultimately 
propagated so the generic demux code in libavformat can create complete 
packets.

> 
> 
>>> 3. How to well test the cbs?
>>
>> Passthrough is the best initial test, by making an h266_metadata bsf (even
>> if it has no options).  The existing test coverage in FATE of CBS is
>> primarily driven by this - for H.26[45], the input streams there are a
>> subset of the H.26[45].1 conformance test streams chosen to cover as much
>> of the header space as possible.
>>
>> I recommend enabling the write support ASAP (by adding the second include
>> of cbs_h266_syntax_template.c), because the double build can shake out
>> other problems too.
>>
>> My original plan is cbs->parser->vvdec decoder->60% comfornace passrate ->
> check in -> h266_metadata.
> are you suggest we implement h266_metadata firstly?
> 
> thanks
> 
> Thanks,
>>
>> - Mark
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
>