/install-tensorflow-on-friendlycore

Primary LanguageShell

Installing Tensorflow 1.14.0 on FriendlyCore/FriendlyDesktop

Here are the packages and utilities your system will have after you follow the instructions in this tutorial:

  • Tensorflow 1.14.0
  • Python 3.5.6

Currently supported boards

  • RK3399
    NanoPC T4
    NanoPC M4
    NanoPC NEO4

Installation

Step1 - Clone this repository on target board

git clone https://github.com/friendlyarm/install-tensorflow-on-friendlycore.git

Step2 - Build python3.5 from source

cd install-tensorflow-on-friendlycore
./01-install-python-3.5.sh

Step3 - Install tensorflow 1.14.0-aarch64

./02-install-tensorflow.sh

Test image recognition sample

git submodule init
git submodule update
cd examples/models/tutorials/image/imagenet
python3 classify_image.py

classify_image.py downloads the trained model from Google’s backend, when the program runs the first time. You'll need about 200 MB of free space available on your disk.
image
The above commands will classify a supplied image of a panda bear.
If the model runs correctly, the script will produce the following output:

giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca (score = 0.89107)
indri, indris, Indri indri, Indri brevicaudatus (score = 0.00779)
lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens (score = 0.00296)
custard apple (score = 0.00147)
earthstar (score = 0.00117)