[FFmpeg-devel] [RFC/PATCH] MV-HEVC decoding

[James Almer]

On 9/13/2024 5:51 PM, Danny Hong wrote:
> Quoting James Almer (2024-09-13 22:19:35)
>> Are you talking about decoding the resulting raw bitstream created from
>> doing stream copy from the source mov? If so, what's happening is
>> probably that the parser is splitting AUs when the second SEI with
>> layer_id == 0 shows up in those non-conformant samples.
>> This doesn't happen if you try to decode directly from the mov as the
>> parser does not attempt to do any packetization then.
> 
> Yes, this is due to hevc_find_frame_end (in libavcodec/hevc/parser.c)
> correctly marking a new AU when prefix NAL with layer ID 0 is detected.  I
> think Apple is incorrectly placing the prefix NAL to precede the VCL NAL
> with layer ID 1.

It's probably not misplaced but mistagged as nuh_layer_id == 0 when it 
should be 1 (Can SEI NALUs be in anything other than the base layer?).
The SEI is User Data Unregistered, and its contents are different than 
the one preceding the base layer VCL NALU. so i guess it's some Apple 
specific metadata that applies to each corresponding view that they 
didn't bother to document.

Can this issue be raised to them?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20240913/f61225f3/attachment.sig>