Dockerfile.cudasamples

1) Install NVidia SDK & download all packages for the Jetson NANO. Do not flash the board from SDK manager.
2) run ./prepare.sh <path_to_sdk_downloads_dir>
3) Build Dockerfile.cudasamples
4) In the app container, start the X display server and run the prebuilt sample apps:
    $ X &
    $ ./clock
    $ ./deviceQuery
    $ ./postProcessGL
    $ ./simpleGL
    $ ./simpleTexture3D
    $ ./smokeParticles

NOTE: CUDA runtime libraries are very large, some even reaching 100-250MB. Therefore not all sample apps have been built in order to keep image size low.
To build more examples please add them to the samples section of the Dockerfile.

If headless GPU computing needs to be performed, without CUDA runtime and without using a display, comment out the path of runtime libs as mentioned in the docker file and remove xorg from the final image.
A GPU API only image can be cut down to less than 400 MB. With CUDA libraries installed the image size goes slighthly over 800 MB whereas an image with video output support - X display server and goes at around 1.3 GB.

Dockerfile.opencv

This dockerfile presents how OpenCV can be built with CUDA libraries inside a container.
a) Perform steps 1) and 2) from above but remove libcudnn debs to improve upload time
b) Build Dockerfile.opencv
c) In the app container run
   $ export DISPLAY=:0
   $ ./example_ximgproc_fourier_descriptors_demo
   $ ./example_ximgproc_paillou_demo corridor.jpg

NOTE: This example builds only a small subset of OpenCV & CUDA modules to keep build time and image size at a low point.
Building specific OpenCV modules is done by selecting them in the build args: cmake -D BUILD_LIST=module1,module2.
Depending on specific usecase, select modules may need to be added as dependencies, otherwise OpenCV build may fail.

In this format the resulting build image is around 1.05 GB.

Dockerfile.gstreamer

This example presents how NVidia gstreamer libraries containing gst elements can be used withing a gstreamer pipeline.

a) Build Dockerfile.gstreamer
b) After application is successfully uploaded to the jetson-nano board, execute in a container:
   $ gst-inspect-1.0 |grep omx
     mx:  nvoverlaysink: OpenMax Video Sink
     omx:  omxvp9enc: OpenMAX VP9 Video Encoder
     omx:  omxvp8enc: OpenMAX VP8 Video Encoder
     omx:  omxh265enc: OpenMAX H.265 Video Encoder
     omx:  omxh264enc: OpenMAX H.264 Video Encoder
     omx:  omxwmvdec: OpenMAX WMV Video Decoder
     omx:  omxmpeg2videodec: OpenMAX MPEG2 Video Decoder
     omx:  omxvp9dec: OpenMAX VP9 Video Decoder
     omx:  omxvp8dec: OpenMAX VP8 Video Decoder
     omx:  omxh265dec: OpenMAX H.265 Video Decoder
     omx:  omxh264dec: OpenMAX H.264 Video Decoder
     omx:  omxmpeg4videodec: OpenMAX MPEG4 Video Decoder
     libav:  avenc_h264_omx: libav OpenMAX IL H.264 video encoder encoder

c) A test pipeline can be constructed like so:
   $ gst-launch-1.0 videotestsrc ! 'video/x-raw, format=(string)I420,width=(int)640, height=(int)480' ! omxh264enc ! 'video/x-h264, stream-format=(string)byte-stream' ! h264parse ! omxh264dec ! nvoverlaysink -e
     Pipeline is PREROLLING ...
     Framerate set to : 30 at NvxVideoEncoderSetParameterNvMMLiteOpen : Block : BlockType = 4
     ===== NVMEDIA: NVENC =====
     NvMMLiteBlockCreate : Block : BlockType = 4
     H264: Profile = 66, Level = 40
     NvMMLiteOpen : Block : BlockType = 261
     NVMEDIA: Reading sys.display-size : status: 6
     NvMMLiteBlockCreate : Block : BlockType = 261
     Allocating new output: 640x480 (x 24), ThumbnailMode = 0
     OPENMAX: HandleNewStreamFormat: 3595: Send OMX_EventPortSettingsChanged: nFrameWidth = 640, nFrameHeight = 480
     Pipeline is PREROLLED ...
     Setting pipeline to PLAYING ...
     New clock: GstSystemClock

d) While the gstreamer pipeline from step c) is running, open a HostOS tab in the dashboard and execute the following line:
   ~# cp $(find /  | grep tegrastats | grep data) /tmp/ && /tmp/tegrastats

   You should see logs similar to this:
   RAM 871/3964MB (lfb 29x4MB) IRAM 0/252kB(lfb 252kB) CPU [25%@307,19%@307,25%@307,15%@307] EMC_FREQ 2%@1600 GR3D_FREQ 0%@76 NVENC 716 NVDEC 716 APE 25 [email protected] [email protected] PMIC@100C [email protected] AO@44C thermal@36C POM_5V_IN 2798/2789 POM_5V_GPU 0/0 POM_5V_CPU 370/378

   NVENC and NVDEC show that the hardware decoders are currently running. If you stop the gst-launch-1.0 application, NVENC and NVDEC will no longer be displayed by tegrastats.