Introduction to the Deep Learning Workbench in the Intel® Distribution of OpenVINO™ toolkit on GNU Linux*

Published: 04/04/2019  

Last Updated: 04/04/2020

workbench page


Operating System

Ubuntu* 16.04, Ubuntu 18.04

Other Linux distributions, such as CentOS* 7, are not validated.

Toolkit Original *.tgz for l_openvino_toolkit
Container Docker* CE (all options require docker)
Storage Space 5 GB
Web Browser

Google Chrome* 72+

Other browsers, like Mozilla Firefox* 65 or Apple Safari* 12, are not validated.

Microsoft Internet Explorer* is not recommended

Skills Familiarity with Linux* commands

The Deep Learning Workbench was introduced with the Intel® Distribution of OpenVINO™ toolkitlkit recently. In this article, we will setup and demo the deep learning workbench.

The instructions are broken up into 6 different steps.

Step Description
1 Install the Deep Learning Workbench
2 Configure - Import Model(s)
3 Configure - Import or Create Dataset
4 Configure - Select Environment
5 Execute

Stage 1: Install the workbench

Note: The workbench included in the Intel® Distribution of OpenVINO™ toolkitlkit 2019 r3 and 2020 r1 require docker*.

Open a terminal. Run the following command:

sudo docker run -p --name workbench --privileged -v /dev/bus/usb:/dev/bus/usb -v /dev/dri:/dev/dri -e PORT=5665 -e PROXY_HOST_ADDRESS= -it openvino/workbench:latest

This command sets up the environment for the DL Workbench Docker* image, builds the Docker image, and runs the Docker container with the DL Workbench.

Once the command is executed, you may see a message similar to:

Proxy server listening on:
new session: ID

Once you verify a session is running, use the browser to access: (localhost)

The DL Workbench Get Started page displays.

workbench page

Step 2: Configure - Import Model(s)

Only one model can be imported at a time with r3 and 2020 r1 workbench. After import, the models are available. However, note the models may take time to import one at a time. Either Upload an Original Model or select a model to import from the Open Model Zoo.

Select the Import Model button.

import model button


import button workbench

Upload Options

import convert

The frameworks available include:

  • Caffe*
  • Apache* MxNet*
  • ONNX*
  • TensorFlow*

import optionsImport from Open Model Zoo

  1. Select Open Model Zoo
  2. Select the model to import: face-detection-retail-004
  • Framework: squeezenet - OpenVINO IR
  • Type: object detection
  • Precision: FP16

face detection selection to import

3. Select the + sign to see more details.

4. The details display.

details for the model to import

5. Select Import

Progress of the import displays under Status.

importing model

The import completes.

model finished importing

3. Configure - Import or Create Dataset

Add a dataset. The options are:

  • Import Local Dataset
  • Autogenerate Dataset in ImageNet format

autogenerate imagenet dataset

Select Autogenerate Dataset in ImageNet format if you do not have a local dataset to import.

The default settings are 100 images at 124x124.

There is an option to select more images. The more images you select to include in the dataset, the longer this will take to autogenerate.

auto generate default settings

The dataset is complete.

autogenerate done

In this example, the squeezenet model will not work with an imagenet dataset, but a caffenet model will.

incompatible dataset and model

Import a caffenet model.

compatible dataset and model - caffe and imagenet

After importing caffenet, select caffenet with an autogenerated dataset.

4. Configure Environment

Select an environment from the list.

  • Core i7
  • GPU
  • Myriad

intel environment

Note: For the purposes of this article, the Up* Xtreme was used and the listed options are available when using the Up Xtreme. Your environment options may vary.

Click GO.

5. Execute

After clicking GO, the results display.


Install DL Workbench

Deep Learning Workbench Security

New Deep Learning Workbench Profiler Provides Performance Insights

Product and Performance Information


Performance varies by use, configuration and other factors. Learn more at