Skip to content

Latest commit

 

History

History
80 lines (52 loc) · 2.95 KB

EmbeddedSystem.md

File metadata and controls

80 lines (52 loc) · 2.95 KB
Raw

How to run on Jetson TX1

Tested with the following package versions:

  • L4T 28.2.0 [ JetPack 3.2 ]
  • Ubuntu 16.04.6 LTS
  • Kernel Version: 4.4.38-tegra
  • CUDA 9.0.252
  • CUDA Architecture: 5.3
  • CUDNN: 7.0.5.15
  • Python: 3.5

Dependencies

  • Tensorflow:

    There isn't a pre-compiled tensorflow version for Jetpack v3.2, but you may install tensorflow 1.14 compiled for Jetpack v3.3, which is available at this link.

  • Keras:

    Keras is a framework that acts as a front-end for tensorflow. To install a version compatible with the tensorflow version previously installed, run the following command:

    $ pip install keras==2.2.4

    This will install the version 2.2.4 of Keras.

  • NumPy:

    To maintain the compatibility between libraries, it might be necessary to update the numpy version. To install the version 1.16.0 (the one installed for the tests), run:

    $ pip install numpy==1.16.0

Running the test

  1. Cloning the repository

    The first step is to clone this repository and download the trained weights. If you already did this, just skip to the step 3. If you didn't, to clone this repository just run this:

    $ git clone https://github.com/LucasNolasco/ScatNILM.git
cd ScatNILM

  2. Downloading the weights

    The weights might be download directly from the following url on a browser. Another easy option is using the gdown program. First of all, it's necessary to install it:

    $ pip3 install gdown

    After the instalation, just run the following command to download the weights:

    $ gdown https://drive.google.com/u/1/uc\?id\=1x1gXS_n9HYSvwDikudOLiPR1NATX3kqv\&export\=download

    This will create a new file named TrainedWeights.zip on the current directory. The final step is to unzip this new file:

    $ unzip TrainedWeights.zip

    After all this, the final result should be a folder called TrainedWeights inside the repository folder (ScatNILM).

  3. Running the test

    To maximize the board performance, these tests assume that the jetson_clocks script is running.

    Inside the repository folder (DeepDFML-NILM), enter the script folder and run the test script. All the code on this repository is designed for Python3, but the tensorflow is compiled for Python2, so this code must be executed using Python2 on Jetson. To run it:

    $ cd script
$ python test_jetson.py

    In order to check the board resource consumption, the package jetson-stats was employed. Install instructions and usage examples are found on its GitHub repository.