pytorch_toolkit

Training Toolbox for PyTorch*

Training Toolbox for PyTorch* provides a convenient environment to train Deep Learning models and convert them using OpenVINO™ Toolkit for optimized inference.

Prerequisites

• Ubuntu* 16.04 / 18.04
• Python* 3.4-3.6
• libturbojpeg
• For Python prerequisites, refer to requirements.txt
• (Optional) OpenVINO™ R3 to export trained models

Quick Start Guide

Setup Training Toolbox for PyTorch

1. Create virtual environment:

   cd /<path_to_working_dir>/training_toolbox/pytorch_toolkit/<model>
   bash init_venv.sh
   

2. Start working:

   . venv/bin/activate
   

   NOTE: If you have installed the OpenVINO™ toolkit after creating a virtual environment, recreate one to install required packages for the Model Optimizer into a single virtual environment.

NOTE: Update several environment variables required to compile and run OpenVINO™ toolkit applications, for details see Install Intel® Distribution of OpenVINO™ toolkit for Linux*.

Models

After installation, you are ready to train your own models, evaluate and use them for prediction.

• Action Recognition
• Face Recognition
• Human Pose Estimation
• Instance Segmentation
• Object Detection
  • Face Detection
  • Person Vehicle Bike Detector
• Segmentation of Thoracic Organs
• Super Resolution

Tools

Tools are intended to perform manipulations with trained models, like compressing models using Quantization-aware training or sparsity.

• Neural Networks Compression Framework