Demo summary included with OpenVINO™ (based on 2020.3)

Introduction

This article is a summary of the demo applications included in the Intel® OpenVINO™ Toolkit v2020.3.

We hope you will find it useful to learn more about the demos that Intel® has made available and see if any of them are suitable for your application.

Reference:

• Intel® demo introduction page
  • https://docs.openvinotoolkit.org/2020.3/_demos_README.html

1. types of demos included with OpenVINO™ and how to use them

The demo applications that come with OpenVINO™ are broadly classified into three types, and are provided in the following folders. (Unrelated folders are omitted.)

This page introduces the demo applications provided as Inference Engine demos.

• OpenVINO™ Demo for Operation Verification ( deployment_tools/demo)
  • Demo with scripts to download models and build applications at once
  • Can be used after OpenVINO™ installation to verify the operation.
• Inference Engine Samples (deployment_tools/inference_engine/samples)
  • Simple coding and can be used to understand how to use the provided APIs.
  • The demo is simple as well.
• Inference Engine Demos (deployment_tools/open_model_zoo/demos)
  • Used as a reference for using models provided by Open Model Zoo (OMZ)
  • Many of them look good as demos, but the code is a bit complicated.

 <Install DIR> 
 | 
 |- deployment_tools -| 
 |- demo 
 | 
 |- inference_engine -| 
 |- samples 
 | 
 |- open_model_zoo -| 
 |- demos 
 

NOTES:

Each demo video introduces "features" based on information disclosed by Intel® and does not maximize "performance". Performance can be verified using the Benchmark App.

• About Benchmark App

2. execution environment

The following is an example of the execution environment using IEI's Flex-BM200 industrial PC and various accelerator products.

• Kit: IEI Flex-BX200
• CPU : Intel® Core™ i7-8700T CPU @ 2.40GHz
• Memory : DDR4 16GB ( 2 x 8GB, 2133MT/s)
• Storage : 150GB (SSD)
• OS : Ubuntu 18.04.3 LTS
• SDK : OpenVINO 2020.3
• Accelerator : Ubuntu 18.04.3 LTS
  • Mustang-V100 (with 8 Intel® Movidius™ Myriad™ X FPs)
  • Mustang-F100 (with Intel® Arria™ 10 GX FPGA)
• Click here for product details
  • https://www.macnica.co.jp/business/semiconductor/manufacturers/iei_integration/ (Japanese)

3. demo list published by Intel® (excerpts)

This is a list of demos that Intel® has made available to the public, focusing on the most in-demand "human-related" demos.

All videos were actually performed using IEI Flex-BX200, Mustang-F100, and Mustang-V100. We have not optimized them for each device, so some may require optimization, and some may not work because they do not support the accelerator in the first place.

Click on the device name at the bottom of each demo to jump to the demo video when run on that device.

Tip:

The input videos used in the demos are free and publicly available from Intel®. Please also use this list.

• List of videos provided by Intel®.
  • https://github.com/intel-iot-devkit/sample-videos

3D Human Pose Estimation Python* Demo Detects a person's skeleton and displays the detected person in 3D. Demo movies (by execution device) FPGA bitstream is 2020-3_PL2_FP16_ELU.aocx				Action Recognition Python* Demo * Source: https://github.com/openvinotoolkit/open_model_zoo/blob/master/models/intel/driver-action-recognition-adas-0002/description/ driver-action-recognition-adas-0002.md (Reference information in still images only as we do not have video data suitable for running the demo)
CPU	GPU	FPGA	HDDL	CPU	GPU	FPGA	HDDL
Crossroad Camera C++ Demo Detects people and displays attribute data of detected people Demo Video (by execution device) FPGA bitstream is 2020-3_PL2_FP11_YoloV3_ELU.aocx				Gaze Estimation C++ Demo Detects faces and infers eyes and their gaze from the detected faces. Demo movies (by execution device) FPGA bitstream is 2020-3_PL2_FP11_TinyYolo_SSD300.aocx
CPU	GPU	FPGA	HDDL	CPU	GPU	FPGA
Human Pose Estimation C++ Demo Detects and displays human skeleton Demo movies (by execution device) FPGA bitstream is 2020-3_PL2_FP11_MobileNet_Clamp.aocx				Interactive Face Detection C++ Demo Infer gender, age, and facial expression from a face. Demo Movie (by execution device) FPGA bitstream is 2020-3_PL2_FP11_MobileNet_Clamp.aocx
CPU(Sync)	GPU(Sync)	FPGA(Sync)	HDDL(Sync)	CPU(Sync)	GPU	FPGA
CPU(ASync)	GPU(ASync)	FPGA(ASync)	HDDL(ASync)
Interactive Face Recognition Demo Interactive Face Recognition recognizes a person based on registered face information. Demo Video (by device)
CPU	GPU
Multi-Channel Human Pose Estimation C++ Demo Detects and displays the human skeleton in multiple videos. Demo Video (by execution device) FPGA bitstream is 2020-3_PL2_FP11_MobileNet_Clamp.aocx				Multi-Channel Face Detection Demos Detects faces in multiple videos. Demo Video (by execution device)
CPU	GPU	FPGA	HDDL	CPU	GPU		HDDL
Pedestrian Tracker C++ Demo Detects a person, assigns an identification number, and displays the track in the camera. Demo Movie (by execution device) FPGA bitstream is 2020-3_PL2_FP11_MobileNet_Clamp.aocx				Security Barrier Camera C++ Demo Detects vehicles and displays their colors. Also displays license plate number if detected. Demo Video (by device)
CPU	GPU	FPGA		CPU	GPU		HDDL
Single Human Pose Estimation Demo Detects a person and infers left and right shoulders, hips, knees, ears, etc. Demo Videos (by execution device)				Smart Classroom C++ Demo Detects people and infers student behavior in a classroom. Demo Video (by execution device) FPGA bitstream is 2020-3_PL2_FP11_SqueezeNet.aocx
CPU	GPU		HDDL	CPU	GPU	FPGA	HDDL
Text Detection C++ Demo Estimates and displays text (alphanumeric characters) in an image. Demo Movie (by execution device) FPGA bitstream is 2020-3_PL2_FP11_MobileNet_Clamp.aocx				Text Spotting Python* Demo Extracts words from an image and displays them in different colors. Demo Video (by execution device) FPGA bitstream is 2020-3_PL2_FP11_AlexNet_GoogleNet_Generic.aocx
CPU	GPU	FPGA	HDDL	CPU FPGA		FPGA
Handwritten Japanese Recognition Python* Demo Detects and estimates handwritten Japanese characters in images. Demonstration movies (by execution device) FPGA bitstream is 2020-3_PL2_FP11_InceptionV1_ResNet_VGG.aocx
CPU	GPU	FPGA	HDDL

Notes:

The script (demo_menu.sh) used in each demo video was created by our company. Please refer to the following links to confirm the arguments and other details on how to execute each demo.

• https://docs.openvinotoolkit.org/2020.3/_demos_README.html

4. summary

As mentioned above, OpenVINO™ provides a variety of functional demo samples. The pre-trained models used in the demos are available for commercial use, so you can add your own functionality by referring to a demo that is similar to your desired process. This will allow users to focus on their own original functions and contribute to product development in a short period of time, so we encourage you to take advantage of this service.

5. reference links

• OpenVINO™ / Accelerator Related Information