gstreamer pipeline unable to find tensorflow version installed in the Ubuntu system

[aks473]

It might be very trivial but i need help. I'm trying to run nnstreamer for the first time , prepared a fresh Ubuntu 18.04, installed gstreamer by following instructions from Gstreamer official site. Then installed tensorflow 1.13.1. Followed instruction for installing nnstreamer using PPA and nnstreamer examples also using PPA.
When i tried to run gstreamer_objectDetection.sh sample , i'm getting an error --> Cannot identify the neural network framework , tensorflow. Even though i've installed it. and i think because of that i'm seeing another error could not link tensorconvertor0 to tensorfilter0

[myungjoo]

You need to let others recreate the issue you are experiencing.
Please show the shortest and simplest code and procedures to recreate your issue.

Please run the utility, nnstreamer-check, and show the result as well as your pipeline and its error/warning messages.

[aks473]

Followed the steps mentioned in here for installing nnstreamer and samples. As you could see in the screenshot i'm still getting error while running the gst-launch-object-detection-tf.sh.
I also have tensorflow=1.13.1 and gstreamer installed in the system.

Basically i have setup a fresh Ubuntu 18.04 with

1. installed tensorflow 1.13 on it
2. installed gstreamer by following instructions at

3) installed nnstreamer (https://nnstreamer.ai/#get-started)

4. installed nnstreamer examples

I'm definitely missing something here, must be very basic
screenshot of printenv

[myungjoo]

We need to check

1. is nnstreamer properly installed?
   • install nnstreamer-util from PPA. It will show related info w/ nnstreamer-check
2. is nnstreamer-tensorflow property built and installed?
   • If you want to use tensorflow of your own, not from nnstreamer PPA, you may need to rebuild nnstreamer-tensorflow depending on the versions and additional features/patches of your tensorflow. Try tensorflow and nnstreamer-tensorflow from nnstreamer ppa first.
3. what is your pipeline? (shortest and simplest that recreates your issue)

[aks473]

Thanks a ton for the propmt response.

Output of nnstreamer-check

I've just started exploring nnstreamer and i wanted it to run with the instructions provided on the site. I have no need for having my own tensorflow version , i thought nnstreamer might need it to run the samples. That's the reason i had installed gstreamer and tensorflow prior to installing nnstreamer.

Our product is using gstreamer/DLStreamer and Openvino ,
so i'm trying the pipelines available under nnstreamer samples like object detection to get a hang of it and decide on how we could leverage nnstreamer

[aks473]

I had reinstalled nnstreamer, it's plugins , nnstreamer-example without on a fresh Ubuntu 18.04 but still facing same issue of couldn't identify the give NN Framework while running gst-launch-object-detection-tf.sh . Didn't install any tensorflow version separately on the system.

[myungjoo]

You don't have nnstreamer-tensorflow installed. With tensorflow subplugin, nnstreamer-check should look like:

...
[Filter]
  tensorflow
    - No custom property found
  python3
    - ${GENERAL_STRING}: There is no key-value pair defined by python3 subplugin. Provide arguments for the given python3 script.
  tensorflow1-lite
    - NumThreads: Number of threads. Set 0 for default behaviors.
    - Delegate: TF-Lite delegation options: {'NNAPI', 'GPU', 'XNNPACK', 'External'}. Do not specify to disable delegation.
    - ExtDelegateLib: Path to external delegate shared library
    - ExtDelegateKeyVal: key/values pairs optional parameters for delegate. Format ExtDelegateKeyVal=key1#value1;key2#value2...
  cpp
    - No custom property found
...

So,

1. install tensorflow and nnstreamer-tensorflow from PPA or (the fastest method. if you have installed nnstreamer via ppa; do $ sudo apt install nnstreamer-tensorflow)
2. build nnstreamer-tensorflow in your system with tensorflow & its developmental files ready. (if you are familiar with build&install from C/C++ sources in Linux)

When you build(rebuild) nnstreamer w/ meson, add -Dtf-support=enabled (refer to meson_options.txt) to ensure that meson fails if it cannot find a proper installation of tensorflow; otherwise, it will simply skip building nnstreamer-tensorflow.