libnvomx.so : H264 encoder : high profile with 422 colorspace support.

Autonomous Machines Jetson & Embedded Systems Jetson TK1
1

We are using the Nvidia hardware h264 video encoding engine on Jetson TK1 , using firmware R21.4.
For our application, we have to set the H264 encoding profile to High (High 4:2:2) because we grab CSI input frames having a UYVY pixel format (Name: YUV422 (UYVY) packed ).

We did some tests with gstreamer-1.0 (ref http://developer.download.nvidia.com/embedded/L4T/r21_Release_v4.0/L4T_Jetson_TK1_Multimedia_User_Guide.pdf) and we are not able to set the h264 encoding profile(libnvomx always return OMX_IndexParamVideoProfileLevelCurrent not implemented).

Some research bring us to this topic https://devtalk.nvidia.com/default/topic/854385/-zjuislab-several-questions-about-video-encoding-on-tegra-k1/ describing the same issue.

On nvidia dev forum, it seems that an update of the libnvomx.so can solve it, and you already provided it.

Can we please get a newer version of the libnvomx.so for Tegra K1 that support H264 high profile encoding ?

Here is the dump of the file /etc/nv_tegra_release containing the SHA1 of the libnvomx.so
root@tegra-ubuntu:~# cat /etc/nv_tegra_release

# R21 (release), REVISION: 4.0, GCID: 5650832, BOARD: ardbeg, EABI: hard, DATE: Thu Jun 25 22:38:59 UTC 2015
3afd51954af5246ad7139897f5f856b9c9748d99 */usr/lib/arm-linux-gnueabihf/tegra/libnvomx.so
2

Any update to that topic ?

Does the HW H.264 encoder support H264 high profile encoding with UYUV 422 input ?

3

Hello, Romary:
For H264 high profile, 21.4 does not support. I will check internally whether it can be included in next release.

For UYVY (not UYUV, I guess) 422 input, it should be converted by nvvidconv before fed to omxh264enc.

br
ChenJian

4

Hello Chen,

Could you confirm that nvvidconv is a closed source gstreamer plugin that only use NEON instruction ?

Could you confirm that nvvidconv doesn’t use any other hardware (that neon) to compute image transformation ?

Regards,

5

Hello, Romary:
nvvidconv is a hardware-accelerated engine for image conversion. The main task of this module is done by hardware, not CPU.

br
Chenjian