How to Decode Video Efficiently

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
#1

Hi,
I need your help to decode h265 video efficiently with minimum processing time.
I need the decoded video output on a single consecutive cpu memory buffer - without gaps between chunks of data (e.g. no pitch).
I work with the multimedia APIs.
If I understand correctly after capture plane deque I need:
a. to convert to pitch linear.
b. since I need a consecutive data without gaps - somehow eliminate the pitch since it is not zero.
c. transfer everything to cpu memory
Looking into the 00_deocde sample I see

  1. a call to NvBufSurf::NvTransform() but measuring the processing time it doesn’t look very efficient (takes some 4-5ms to transform a 1080x1920 video)
  2. and a call to dump_dmabuf()

My question is if there is an alternative more efficient way I could employ?

I work on AGX Xavier
my nv_tegra release is:
cat /etc/nv_tegra_release
R34 (release), REVISION: 1.0, GCID: 30102743, BOARD: t186ref, EABI: aarch64, DATE: Wed Apr 6 19:11:41 UTC 2022

Thanks,
Idof

#3

Hi,
Are you able to try Jetpack 4.6.1? It is production release for AGX Xavier.

r34.1 is Jetpack 5.0 DP which is developer preview.

#5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.