https://github.com/yas-sim/openvino-visual-performance-demo

Intel OpenVINO visual performance demo - Supports classification models.
https://github.com/yas-sim/openvino-visual-performance-demo

deep-learning demo glut image-classification inference intel opengl openvino performance-visualization python

Last synced: about 2 hours ago
JSON representation

Intel OpenVINO visual performance demo - Supports classification models.

• Host: GitHub
• URL: https://github.com/yas-sim/openvino-visual-performance-demo
• Owner: yas-sim
• Created: 2020-09-24T14:13:35.000Z (about 4 years ago)
• Default Branch: master
• Last Pushed: 2021-06-11T04:53:32.000Z (over 3 years ago)
• Last Synced: 2024-04-23T00:09:54.101Z (7 months ago)
• Topics: deep-learning, demo, glut, image-classification, inference, intel, opengl, openvino, performance-visualization, python
• Language: Python
• Homepage:
• Size: 15 MB
• Stars: 9
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# OpenVINO Visual Performance Demo Program

## Overview

This is an OpenVINO performance demo program. This program demonstrates inferencing performance visually on the screen so that the audience can get the performance of the system and OpenVINO intuitively.  

The proram supports simple image classification models models.   

**NOTICE**  

The program has been updated for performance up. The previous version has performance bottle neck caused by OpenCV. The new version replaced some OpenCV portion with OpenGL. Also, some features such as SSD model support is dropped to simplify the program.  

OpenVINOを使ったパフォーマンスデモプログラムです。推論性能を目で見てわかる形でデモしますので異なるシステムやデバイス間でのパフォーマンスの違いをわかりやすくデモできます。  

このプログラムはいわゆる画像分類モデルのみをサポートしています。  

最初のバージョンはOpenCVに起因するパフォーマンス問題があったのですが、一部をOpenGL化することで大幅な高速化を達成しました。また、以前のバージョンはSSDなどのモデルもサポートしていましたが遅くなるので止めました。  

### You can use this program to demonstrate side-by-side performance comparison demo of OpenVINO.  

![video](./resources/visual-demo.gif)

## Prerequisites

- OpenVINO 2020.4

- OpenGL, GLUT (FreeGLUT)  

`(Ubuntu) sudo apt install python-opengl freeglut3 freeglut3-dev`   

- Some Python modules  

` pip install pyyaml,numpy,opencv-python,PyOpenGL,PyOpenGL_accelerate`

- Image files for inferencing  

  - Annotation file is not required.

  - Place image files in a directory and specify the directory in the YAML configration file.

- Image classification IR model such as ResNet-50  

**NOTICE**  

- The program requires GLUT. If you are using Windows, maybe you need to install GLUT separately to enable it.  

## Configuration

Benchmark configuration can be defined in a YAML file. Create your own configuration by reffering to the `default.yml` as an example.  

## How to run

1. Create or modify YAML configuration file

 - The YAML configuration file contains the parameters for performance demo

 - `default.yml` will be used when no configuration file is given.

2. Run the performance demo script  

`python visual-demo.py -c `

## Tested environment

- OpenVINO 2020.4, 2021.1

- Windows 10

- Ubuntu 18.04

## Update history

 - 07-10-2020: Separated benchmark main loop to a thread. Screen update and control logic is still remain in the main thread. This minimizes the performance impact from screen update.  

 - 16-10-2020: Replaced OpenCV codes with OpenGL. Droped support for SSD models.