GitLab

But use avcodec_alloc_frame()/avcodec_free_frame()
See https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=699722 for reference.

If a device is capable of YUYV non-emulated format and
YUV420P emulated (cpu color conversion), then pick the
non emulated one.
In the worst case (encoder with no YUYV support), pixconv
will be used to convert YUYV -> YUV420P.
In the best case (encoder supports YUYV input), we skip a
costly color-conversion step.

Allow these filter to specify if they support non-YUV420P
format (e.g: YUYV). If true, feed them with this format instead
of using pixconv.

If some incomplete VP8 frame is left after an iteration of vp8rtpfmt_unpacker_process, keep them for next iteration.

disconnect callbacks from rtp session when AudioStream is destroyed, otherwise they remain in the rtp session, which is kept for further updates of the session (reINVITE)

Force VP8 keyframe on encoder preprocess only when AVPF is enabled + initialize video starter only on first encoder preprocess.

This mechanism is now handled in oRTP by using the RTCP send algorithm defined in the RFC 3550.

Generate VP8 keyframe when an invalid frame is reported and no reference frame has been acknowledged.

If a VP8 reference frame is not acknowledged, send a reference frame of the same time the following time one is sent.

When a VP8 frame is notified as invalid and is newer than the last reference frame, stop referencing the last frame.

Reference VP8 last frame when encoding a new one unless the frame being encoded is a reference frame.

On VP8 decoding error (when sending PLI or SLI) wait for next reference frame before continuing decoding.