opencv-pipeline

A tool to visualize a series of computer vision operations on an image.

Overview

I wrote this tool to debug and visualize OpenCV pipelines. A pipeline consists of a series of OpenCV operations (known in this project as 'filters'.) Each filter accepts a single OpenCV image, performs some operation, and returns the updated OpenCV matrix.

The pipline follows a simple sequence, as follows:

1. It is provided an original source image.
2. It passes the image to the first filter in the pipeline.
3. The first filter returns the modified image to the pipeline
4. The pipeline passes this image on to the next filter, and so on.

The app has the following features:

• Visualize both the final image and individual internal images.
• Dynamically enable and disable individual filters.
• Automatically disable filters that throw exceptions.
• Easy-to-use JSON description of the pipeline.

The following illustrates key parts of the UI.

Installing

Make sure you have Python 3.6+ installed. Run the following command in your local shell:

pip install opencv-python kivy

Running

A quick start is the following command:

python3 main.py

The following parameters are available.

optional arguments:
   -h, --help                          show this help message and exit
   -i PATH, --image PATH               path to input image or video
   -p PIPELINE, --pipeline PIPELINE    path to json describing pipeline

JSON schema

The JSON file supports the following tags:

tag	description	required?
filter	The name of the class in the opencv-filters.py file	Required
params	Filter-specific parameters used to set initial values	Required if there are any params
enabled	If false, the filter becomes a NOP	Optional
output_id	Filters declared below this filter can use the 'output_id' as an input	Optional
name	This is used as the display name; if not provided, the python class name is used	Optional
input	Specifies the 'output_id' of the filter to use as the input image; if this is not specified, the filter uses the image produced by the filter immediately above it in the pipeline.	Optional
inputs	A list of 'output_id' values which will be passed as a list to the input of the filter; useful for BitwiseAnd/Or filters	Optional, but required to join parallel branches of the pipeline.

To Do

• Add drag-n-drop (perhaps with a platform more suited to this).
• Add Library pane with list of supported filters.
• Add open/save functionality.