Porting pretrained models to OpenVINO

Install OpenVINO following the instructions: https://software.intel.com/en-us/openvino-toolkit

Setup a virtual environment with python3 and its pip module, or with conda (from Miniconda or Anaconda). Then install packages through the specific package list based on the configuration of the machine.

This is the virtuelenv with pip example.

$ mkvirtualenv env -p python3
(env)$ pip install -r package_list_pip.txt

Or using conda, where the following is an example of environment set with CUDA 10.

$ conda env create -f package_list_conda_py37_cu10.yml
$ conda activate py37-torch110-cu10

Specify a model name (inception_v3 as example here) and run the line.

(env)$ python pytorch_onnx_openvino.py --model_name inception_v3

Great appreciation towards the following GitHub repos who really enable the possibilities to easily use the pretrained models.

• @Cadene for https://github.com/Cadene/pretrained-models.pytorch
• @lukemelas for https://github.com/lukemelas/EfficientNet-PyTorch

Great appreciation towards the THOP tool developed by:

• @Lyken17 for https://github.com/Lyken17/pytorch-OpCounter