Habana® Gaudi® Processor Training and Inference using OpenVINO™ Toolkit for U-Net 2D Model

ID 759698
Updated 12/2/2022
Version 1.0




Develop a solution using an end-to-end computer vision workflow with training on Habana® Gaudi®, post-training quantization using Post-training Optimization Tool (POT) and Inference using OpenVINO™ toolkit.

Select Configure & Download to download the reference implementation and the software listed below.   

Configure & Download

Icon representing AI

  • Time to Complete: 25 - 30 minutes
  • Programming Language: Python*
  • Available Software: OpenVINO™ toolkit, Docker*, Helm*, Kubernetes*

Target System Requirements 

Training on Amazon Web Services* (AWS) EC2 DL1 Instance 

  • Ubuntu* 20.04
  • Intel® Xeon® Platinum 8275CL processor @ 3.00GHz (96 vCPUs)
  • HPUs: 8
  • Usable RAM: 784 GB
  • Disk Size: 500 GB


Device configuration for the cluster. One node Kubernetes* cluster has comparable configurations to the below:

  • Ubuntu 20.04
  • Intel® Xeon® Platinum 8375C processor @ 2.90GHz (8 vCPUs)
  • Usable RAM: 32 GB
  • Disk Size: 50 GB

How It Works

The repository contains the model scripts and recipe for training a U-Net 2D model to achieve state of the art accuracy using Image Segmentation with Medical Decathlon dataset and followed by inferencing with OpenVINO™ toolkit on Intel® hardware.

This AI workflow demonstrates the following:

  • U-Net 2D model training using Amazon EC2 DL1 instances which uses Gaudi® Processor from Habana® Labs (an Intel® company).
  • U-Net 2D model optimization and inference using OpenVINO™ toolkit on Amazon M6i Intel® CPU instances powered by 3rd Generation Intel® Xeon® Scalable processors (code named Ice Lake).

Diagram showing the RI in the ecosystem.


This reference implementation provides an AWS* cloud-based generic AI workflow, which showcases U-Net-2D model-based image segmentation with medical decathlon dataset.

The reference implementation is available for use by Docker containers and Helm chart.

The architecture is represented by a complex block diagram.
Figure 1: Architecture Diagram



Get Started


Follow the steps on GitHub* to install the prerequisites.

Install the Reference Implementation

Select Configure & Download to download the reference implementation.

Configure & Download 

Train a Model 

Follow the steps on GitHub to train a model using Habana® Gaudi® Processor on AWS*.

You can choose to run training using Docker* containers or with Helm* chart using Kubernetes*.


Run Optimization and Inference on a Model 

Follow the steps on GitHub to perform optimization and inference using OpenVINO™ toolkit.

You can choose to run inference using Docker* containers or with Helm* chart using Kubernetes*.


Learn More

To continue learning, see the following guides and software resources: