pyrebel

A pure python library that implements abstraction of data.

Installation

From PyPI

python3 -m pip install --upgrade pyrebel

From source

git clone https://github.com/ps-nithin/pyrebel
cd pyrebel
python3 -m pip install .

Running demo programs

Demo programs are located in 'demo/' directory.
cd demo/

1. Image abstraction demo

Usage:
python3 pyrebel_main.py --input <filename.png>
Optional arguments
--abs_threshold <value> Selects the threshold of abstraction. (Defaults to -1, which sets per blob threshold based on the size of the blob.)

For example,
python3 pyrebel_main.py --input images/abc.png --abs_threshold 10
The output is written to 'output.png'

To write layers to file use the following,
python3 pyrebel_main_layers.py --input images/abc.png
This will save the files as layer-3.png, layer-4.png etc.

2. Edge detection demo

This is a demo of edge detection achieved using data abstraction.
Usage:
python3 pyrebel_main_edge.py --input <filename>

For example,
python3 pyrebel_main_edge.py --input images/wildlife.jpg

The output is written to 'output.png'. Below is a sample input image,


Below is the output image,



See how edge detection works here

3. 2D sketch demo

This is a demo of 2D sketch formation using data abstraction.
Usage:
python3 pyrebel_main_vision.py --input <filename>

Optional arguments for tweaking the result,

1. --edge_threshold <value> Selects the threshold of edge detection.(Defaults to 5)
2. --abs_threshold <value> Selects the threshold of output abstraction. (Defaults to 10)
3. --bound_threshold <value> Selects the threshold of boundary size. (Defaults to 100)
   
   

For example,
python3 pyrebel_main_vision.py --input images/lotus.jpg

Below is a sample input image,


Below is the output image,

4. Abstract painting

This is a demo of abstract painting using data abstraction. The output of edge detection is painted to obtain the desired output.
Usage:
python3 pyrebel_main_paint.py --input <filename>

Optional arguments for tweaking the result,

1. --edge_threshold <value> Selects the threshold of edge detection. (Defaults to 10).
2. --paint_threshold <value> Selects the threshold of painting. (Defaults to 5).
3. --block_threshold <value> Selects the threshold of block size. (Defaults to 20).
4. --scale_factor <value> Selects the scale of the output image. (Defaults to 1.0).
   
   For example,
   Running python3 pyrebel_main_paint.py --input images/elephant.jpg --edge_threshold 10 --block_threshold 50 --paint_threshold 2 --scale_factor 0.5
   
   Below is the sample input image,
   
   
   Below is the output image,
   
   

5. Abstract focusing

This is a demo of abstract focusing done using data abstraction. It tries to mimic eye movements / saccades / glance.
Usage:
python3 pyrebel_main_edge_join_pan.py --input images/lotus.jpg

You can use keyboard shortcuts to drive focus and movements. Arrow keys for movements, plus / minus keys for focus and t / shift+t for edge threshold.

Optional arguments,

1. --camera 1 Selects camera as input source. (Input csi_pipeline / usb_pipeline to Gst.parse_launch() as required).
2. --output_resolution <value> Selects the output resolution. (Defaults to 600).
   

The output is written to 'output_joined.png'.

6. Pattern recognition demo

This is a demo of pattern recognition achieved using data abstraction.

1. Learning
   Usage: python3 pyrebel_main_learn_multiple.py --learn /path/to/image/directory/
   For example running python3 pyrebel_main_learn_multiple.py --learn images/train-hand/ learns all the images in the directory and links the filename with the signatures.
   
   
2. Recognition
   Usage: python3 pyrebel_main_learn_multiple.py --recognize <filename>
   For example running python3 pyrebel_main_learn_multiple.py --recognize images/recognize.png displays the symbols recognized in the file 'images/recognize.png'.
   
   

Optional arguments,

1. --abs_threshold <value> Selects the threshold of abstraction. (Defaults to -1, which sets per blob threshold based on the size of the blob.)
   

The output is written to 'output_text.png'. To reset the knowledge base just delete file 'know_base.pkl' in the current working directory.
See how the program learns and recognizes patterns here

6.1 Pattern recognition with focusing

This is a demo of pattern recognition combined with focusing.
Usage:
python3 pyrebel_main_learn_multiple_edge_pan.py --camera 1

You can use keyboard shortcuts to drive focus, movements and activate learning mode. Arrow keys for movements, plus / minus keys for focus, t / shift+t for edge threshold and l key for activating learning mode. On activating learning mode you can see the learned patterns in 'inside.png'.

Optional arguments,

1. --input <filename> Selects file as input source.
2. --output_resolution <value> Selects the output resolution. (Defaults to 600).
   

(Note: Make sure that the dimensions of the input image (camera or file) is greater than --output_resolution). The output is written to 'output_text.png'.

7. Abstract motion

This is a demo of obtaining abstract motion data [[angle1,distance1],[angle2,distance2],...] from boundaries.
Usage:
python3 pyrebel_main_motion.py --input images/shapes/rectangle.png

Docs here

Read more here

Let the data shine!