A benchmark to help the decision making process over the Nvidia's Jetson TX2 platform. It's designed to serve as a general purpouse benchmark, being easily customizable for adding and adapting new models to the algorithm.
To start with, it is needed to install all the dependencies. The script has the requirement of using it with python 3.5 or superior.
After installing the python interpreter, it could be as simple as:
pip install -r requirements.txt
After the installation of initial packages, execute:
cd src/ python benchmark.py
The algorithm is responsible by downloading all the needed datasets
and/or models already defined under the
src/networks.py file. If
you want to add your own custom models, generate a tensorflow frozen
graph and host it under this file.
Tensorflow GPU: To install a specific version of Tensorflow, if the jetpack installation failed by some reason, execute:
sudo pip3 install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v$JP_VERSION tensorflow-gpu==$TF_VERSION+nv$NV_VERSION
JP_VERSION is the JetPack version of the specific tensorflow