jetson-ffmpeg
jetson-ffmpeg copied to clipboard
nvmpi decoder fails with some RTSP streams
Trying to use nvmpi with ffmpeg 4.2 -- using commit 3dff99c3c2ef280209244d9bae5771997d4aa770 from this repository.
It seems to be very finicky with the bitstream it attempts to consume. For example:
ffmpeg/bin/ffmpeg -c:v h264_nvmpi -i rtsp://127.0.0.1/short.mkv -f null - - this command fails to decode RTSP stream served by live555 (frames are never generated)
The stream is fine (a different decoder is able to process it):
ffmpeg/bin/ffmpeg -c:v h264 -i rtsp://127.0.0.1/short.mkv -f null - - this command decodes it without issue
The problem isn't exclusive to live555 (we've seen it fail with ffmpeg-served video, albeit not as often), but this is the easiest way to reproduce the problem.
I'm attaching live555.zip
Just running ./live555MediaServer & and one of the commands above should reproduce the issue.
I've the same problem: if I connect to a rtsp stream from a live camera is all ok, but if I connect to a rtsp media server streaming a video in loop it doesn't work: no frame are decoded. I'm using ffmpeg 4.2 and rtsp-simple-server (https://github.com/aler9/rtsp-simple-server) as server. If I launch ffmpeg with the debug loglevel I see this output repeatedly coming till I ctrl-c:
cur_dts is invalid st:0 (0) [init:0 i_done:0 finish:0] (this is harmless if it occurs once at the start per stream)
Thanks for the awesome project.
The problem seems to be not with this wrapper, but the actual nvidia implementation. I'm seeing the same behavior if running with nvidia-supplied ffmpeg on Jetson. Tracking it here - https://forums.developer.nvidia.com/t/nvmpi-decoder-doesnt-work/210993/7
Can you elaborate on 'nvidia-supplied ffmpeg on Jetson' ? From what I've understood, the entire reason why jetson-ffmpeg exists is because nvidia doesn't provide or support ffmpeg with hardware acceleration.
@grantthomas - see https://forums.developer.nvidia.com/t/the-nvidia-ffmpeg-package-supports-hardware-accelerated-decode-on-jetson-platforms/123694?page=2
Interesting, thank you for the link @w3sip