berndpfrommer
commented
6 months ago Is the libnvcuvid installed? https://www.reddit.com/r/jellyfin/comments/10cdxhw/comment/j4fgsbx
On Wed, Feb 7, 2024, 3:55 PM Jorge Miarnau @.***> wrote:
I installed the ros-iron-ffmpeg-image-transport package via apt on a Jetson Orin NX (Publisher) and a Jetson Orin AGX (Subscriber) and get an error when trying to decode the ffmpeg topic without explicitly using hardware acceleration:
Publisher - Orin NX running Jetpack 6 (Ubuntu 22) works ok:
ros2 run image_transport republish raw in:=/amr/infrared_0/image_raw ffmpeg out/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg
[WARN] [1707337736.418144200] [rcl]: Found remap rule 'in:=/amr/infrared_0/image_raw'. This syntax is deprecated. Use '--ros-args --remap in:=/amr/infrared_0/image_raw' instead. [WARN] [1707337736.418209354] [rcl]: Found remap rule 'out/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg'. This syntax is deprecated. Use '--ros-args --remap out/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg' instead. [INFO] [1707337736.674866902] [FFMPEGEncoder]: enc: libx264 prof: preset: [INFO] [1707337736.674984091] [FFMPEGEncoder]: qmax: 10 bitrate: 8242880 gop: 15 [WARN] [1707337737.385677712] [FFMPEGEncoder]: horiz res must be multiple of 32! [libx264 @ 0xaaab16e81130] using cpu capabilities: ARMv8 NEON [libx264 @ 0xaaab16e81130] profile High, level 3.2, 4:2:0, 8-bit [INFO] [1707337737.390063329] [FFMPEGEncoder]: opened codec: libx264
Subscriber - Orin AGX running Jetpack 6 (Ubuntu 22) throws an error:
ros2 run image_transport republish ffmpeg in/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg raw out:=/amr/infrared_0/image_raw/ffmpeg_uncompressed
[WARN] [1707338504.776305805] [rcl]: Found remap rule 'in/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg'. This syntax is deprecated. Use '--ros-args --remap in/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg' instead. [WARN] [1707338504.776384174] [rcl]: Found remap rule 'out:=/amr/infrared_0/image_raw/ffmpeg_uncompressed'. This syntax is deprecated. Use '--ros-args --remap out:=/amr/infrared_0/image_raw/ffmpeg_uncompressed' instead. [INFO] [1707338506.496957587] [FFMPEGDecoder]: decoding with h264 [h264 @ 0xaaaaf0517390] Cannot load libnvcuvid.so.1 [h264 @ 0xaaaaf0517390] Failed loading nvcuvid. [h264 @ 0xaaaaf0517390] Failed setup for format cuda: hwaccel initialisation returned error. Failed to get HW surface format. [h264 @ 0xaaaaf0517390] decode_slice_header error [h264 @ 0xaaaaf0517390] no frame! [WARN] [1707338506.499019759] [FFMPEGDecoder]: send_packet failed for pts: 11503
I have also tried
ros2 run image_transport republish ffmpeg in/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg raw out:=/amr/infrared_0/image_raw/ffmpeg_uncompressed --ros-args -p "ffmpeg_image_transport.map.hevc_nvenc:=libx264"
and
ros2 run image_transport republish ffmpeg in/ffmpeg:=/amr/infrared_0/image_raw/ffmpeg raw out:=/amr/infrared_0/image_raw/ffmpeg_uncompressed --ros-args -p "ffmpeg_image_transport.map.libx264:=h264"
as suggested in another issue but I get the same. I think hardware accelerated ffmpeg is not fully supported on jetson out of the box but I should be able to run this without hardware acceleration?
— Reply to this email directly, view it on GitHub https://github.com/ros-misc-utilities/ffmpeg_image_transport/issues/27, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABPLK2SQWSKHWC53QPVWOHTYSPS3NAVCNFSM6AAAAABC6QI3NSVHI2DSMVQWIX3LMV43ASLTON2WKOZSGEZDGOBXGE2DONQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>
jmm-slamcore
I installed the
ros-iron-ffmpeg-image-transportpackage via apt on a Jetson Orin NX (Publisher) and a Jetson Orin AGX (Subscriber) and get an error when trying to decode the ffmpeg topic without explicitly using hardware acceleration. The images come from a RealSense with 848x480 resolution.Publisher - Orin NX running Jetpack 6 (Ubuntu 22) works ok:
Subscriber - Orin AGX running Jetpack 6 (Ubuntu 22) throws an error:
I have also tried
and
as suggested in another issue but I get the same. I think hardware accelerated ffmpeg is not fully supported on jetson out of the box but I should be able to run this without hardware acceleration?