Hello.
I have faced the problem with "incorrectly" oriented video recordings. As WebRTC users may know, the orientation of a video is encoded in the special RTP header that is described at SDP (in the line like "a=extmap:13 urn:3gpp:video-orientation"), so a receiving device can rotate an incoming video stream if needed. In meantime, Janus records its .mjr`s "as is", without any rotations - and that`s great because it makes the recording process high-performant. We still can find out the rotation of a video stream bypassing -v {$HEADER_NUM} flag to janus-pp-rec util and parsing the textual output. Furthermore, tools like ffmpeg claim that they will rotate your videos by default (-autorotate flag is always on), so it sounds like one can convert .mjr to, say, .webm and then autorotate it to "normal" orientation. But it doesn`t work, even when I use the -v flag with janus-pp-rec. The following code doesn`t rotate the video:
janus-pp-rec -v 13 src.mjr out.webm
ffmpeg -i out.webm trgt.webm
even if janus-pp-rec -v 13 -p src.mjr prints "Video rotation: 90 degrees".
So my question is: should janus-pp-rec add some secret sauce to the resulting .webm metadata in the case of rotated videos? If shouldn`t, why? Is the described situation a bug or the behavior isn`t implemented intentionally? Imho, .webm with orientation metadata could be very convenient for users.
P.S. I saw the solution with "videoorient_ext = false" in the videoroom config, but I use the videocall plugin, and (what's more important) the solution with modifying .webm metadata (if it is possible) seems more universal.