trackpipe

This project is an image processing pipeline that allows you to create reusable transformations and compose them into individual windows to see intermediate output. You can also run a single set of transforms on multiple images simultaneously.

This project is based on the work done by Bob Kerner, here, so kudos to Bob!

Demo

Note: This code has not been tested against python 2.7

OpenCV-Playground

trackpipe was a precursor to opencv-pg - The OpenCV Playground. It's a more full fledged and interactive platform for designing interactive pipelines and testing out built in opencv functions.

Installation

pip install git+https://github.com/geoff-va/trackpipe.git

or if you've cloned the repo: pip install -e trackpipe/

Usage

Sequenced Pipeline

Create individual Transforms once, add them into Windows and reuse/reorder them throughout your pipeline.

• Subclass pipeline.Transform
• Add Param class attributes for each trackbar param you'd like
• Add a draw(self, img) method that performs the transform, accessing your params as self.your_param.value
• Optionally add a compute_values(self) function that further modifies either your self.your_param.value or other instance attributes you've created

class GaussianBlur(Transform):
    """Performs a Gaussian Blur"""
    size = Param(_max=100, adjust=make_odd)
    sigma = Param(label='Sigma', _max=10)

    def draw(self, img):
        return cv2.GaussianBlur(
            img,
            (self.size.value, self.size.value),
            self.sigma.value
        )

Create one of these for each operation you'd like to perform (Convert Color, inRange filter, MedianBlur, HoughLines, Load Video Frame, etc), then compose them into windows (or just a single list):

Multiple Windows

w1 = Window([LoadImage('your_file.png'), Transform1(), ...])
w2 = Window([Transform2(), Transform3(), ...])

pipeline.run_pipe([w1, w2])  # no img param; can use a transform to load and return your image

Single Window

img = cv2.imread('your_file.png')
transforms = [Transform1(), Transform2(), ...]
pipeline.run_pipe(transforms, img)  # Can also pass a loaded image to `run_pipe`

If no name parameter is passed to the Window constructor, it defaults to Step N. Otherwise, you can name the windows if you'd like.

A few simple transforms are included.

If you override the __init__ method, you must call super().__init__() first.

Parallel Pipeline

You can apply a single sequence of transforms to multiple images simultaneously with run_parallel_pipe.

You supply the list of transforms and list of images. The first image will be the master window with trackbar controls, and those trackbars will control the master and all the slaves at the same time.

from pipeline import run_parallel_pipe

images = [
    'path1.jpg',
    'path2.jpg',
]

transforms = [
    YourTransform1(),
    YourTransform2(),
    YourTransform3()
]

run_parallel_pipe(transforms, images)

Examples

If you clone the repo, there are two examples in example.py. To run:

• python example.py --example 1
• python example.py --example 2

Example 1 is a reproduciton of Bob's trackbar example Example 2 is a file load, color convert and two gaussian blurs in successive windows to demonstrate the pipeline