https://github.com/kiritigowda/pythonopenvx

https://github.com/kiritigowda/pythonopenvx

Last synced: 3 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/kiritigowda/pythonopenvx
• Owner: kiritigowda
• Created: 2019-12-19T05:22:48.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2016-10-26T11:05:37.000Z (over 8 years ago)
• Last Synced: 2025-01-14T07:15:12.247Z (4 months ago)
• Size: 6.77 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.txt

Awesome Lists containing this project

README

        
OpenVX based PyOpenVX library:

There are two ways in which PyOpenVX library can be deployed. The first one is by installing a virtual machine, which comprised of compile code and samples. The second one is by manually compile the necessary files.

********************************************************

*               Using a virtual machine                *

********************************************************

We developed a virtual machine with all of the needed packages already installed and compiled. The virtual machine can be download from: 

https://www.dropbox.com/s/0ob5hpmszkkqqda/Ubuntu_PyVX.ova?dl=0

The virtual machine is a VirtualBox virtual machine, in order to run it you need to download VirtualBox from https://www.virtualbox.org/wiki/Downloads, and import our virtual machine into it. If a user name/password is required, you can use: 

User name: ori

Password: 123

Tutorial:

a.	Open VirtualBox

b.	File->Import Application

c.	Choose the virtual machine image ('.ovx' file)

d.	Click 'Next' and then click 'Import'

Run the samples from the VM:

a.	Open shell

b.	Go to /home/ori/Desktop/samples

c.	Run one of the samples:

a.	./sobel_graph.py

b.	./EqualizationGraph.py

The sample code is in the samples folder.

********************************************************

*               Building on a new machine              *

********************************************************

This PyVX extension tested on Ubuntu 16.04.

Prerequisites:

1)	Cmake 2.1 or above (can be download using `sudo apt-get install cmake`)

2)	Download Python 2.7x https://www.python.org/ftp/python/2.7.12/

3)	Download the OpenVX implementation- such as OpenVX sample implementation from: https://www.khronos.org/registry/vx/sample/openvx_sample_1.0.1.tar.bz2 

	a.	Build the OpenVX implantation:

		i.	Unzip the openvx_sample 

		ii.	Run from shell: python Build.py --os=Linux --arch=64

4)	Python libraries: numpy, cffi, PyVX

	a.	Download packages - run from shell: 

		i.	pip install cffi

		ii.	pip install numpy

	b.	Install the PyVX module http://pyvx.readthedocs.io/en/rewrite/ 

		i.	git clone https://github.com/hakanardo/pyvx.git

		ii.	cd pyvx

		iii. git checkout rewrite

		iv.	sudo python setup.py install

		v.	sudo python -mpyvx.build_cbackend --default pyvx /path/to/openvx/install/

5)	Download the Pythonic library from github: https://github.com/NBEL-lab/PythonOpenVX/

6)	Import the Pythonic library and write your App!

Recommended IDE:

We used PyCharm by JetBrains and it was great, very supportive and convenient IDE. 

Link for download: https://www.jetbrains.com/pycharm/

We suggest before trying to run sophisticate graph to start from a single node graph and extend it node by node.

Running:

Import to your python source code the Pythonic library.

Create your OpenVX graph in one of the two ways:

	1)	Create context and graph as regular instance and add the images and the nodes.

	2)	Create a graph as part of with statement.

	

Code Example:

	from pythonic import *

	def sobel_graph(input):

		width = 100

		height = 100

		

		# The OpenVX Context and Graph are created as part of the ‘with’ statement 

		with Graph(verify=True) as g:

			

			# Creating an OpenVX image	

			out = Image(g.context, width, height, Color.VX_DF_IMAGE_RGB)		

			y = ChannelExtractNode(g, input, Channel.VX_CHANNEL_Y)

			

			#Creating and OpenVX Sobel node in the graph

			gx, gy = Sobel3x3Node(g, y)	

			

			#Creating and OpenVX Magnitude node in the graph

			mag = MagnitudeNode(g, gx, gy)	

			#Creating an OpenVX scalar object

			shift = Scalar(g.context, Data_type.VX_TYPE_INT32, 0)

			#Creating an OpenVX conversion node 	

			converted = ConvertDepthNode(g, mag, Policy.VX_CONVERT_POLICY_WRAP, color=Color.VX_DF_IMAGE_U8, scalar=shift)	

			

			# Creating an OpenVX channel combine node

			ChannelCombineNode(g, converted, converted, converted, output=out)	

		

		#Process the OpenVX graph we created

		g.vxProcessGraph()	

	return out