Hello all,
I am reaching out because the company I work for is using FFmpeg to perform
a variety of video processing tasks and we are in need of some assistance
optimizing the performance of some of these tasks.
Some background + platform specifications:
- Currently running a custom build of ffmpeg version 5.1
- Mainly so we can enable hardware encoder/decoder support via
jetson-ffmpeg <https://github.com/jocover/jetson-ffmpeg> and NVIDIA
GPU support.
- Jetson Nano is the platform we are currently targeting (with hopes of
upgrading to the new Orin series).
Our use case involves reading two video streams (mp4) -> remapping the
frames -> doing some overlays -> combining them into one 1920x1080 output
video.
However, we are not getting anywhere near the performance that is needed to
make this viable in a production environment.
On the Nano, the processing runs at approximately 1/6 * frame rate, and on
my Intel desktop, it runs much faster at 6x frame rate.
I have done some benchmarking (timing each stage of the filter
individually) to see what the bottleneck might be but it is still not clear
to me.
I also tried moving some of the processing over to the GPU; however, since
all of the filters don't have CUDA implementations, I believe the
copying of memory from RAM -> GPU in multiple stages actually caused the
pipeline to run slower.
If this problem seems interesting to anyone, please reach out to the email
below.
My company can gladly provide the platform and software we are using so you
can effectively assist us.
All the best,
--
Cole Richardson
Software Engineer | Edge AI, LLC
crichard...@edgeaisolutions.com
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
