[FFmpeg-devel] [PATCH 2/2 v2] avcodec/cbs_av1: keep separate reference frame state for reading and writing

[James Almer]

On 11/15/2019 9:42 PM, James Almer wrote:
> In scearios where a Temporal Unit is written right after reading it using the same
> CBS context (av1_metadata, av1_frame_merge, etc), the reference frame state used
> by the writer must not be the state that's the result of the reader having already
> parsed the current frame in question.
> 
> This fixes writing frames of type Switch.
> 
> Signed-off-by: James Almer <jamrial at gmail.com>
> ---
> This may be needed for other modules as well. I see at least VP9 and h264/hevc
> keeping some sort of state.
> 
> Alternatively, we could just require a given CBS context to be used to either read
> or write, but not both.
> 
> v2: Simpler version

Here's a sample using Switch frames:
https://code.videolan.org/videolan/dav1d-test-data/raw/658e23584d5add2929abe14b3592073fc711e768/8-bit/sframe/autostitch-480p-240p-160p-10s.ivf