(Py)ImGui for Data Science

A set of utilities for data science using python, imgui, numpy and opencv.

• Free software: Apache Software License 2.0
• Documentation: https://imgui_datascience.readthedocs.io.
• Demo:

Compatible with python 2 and python3.

Acknowledgments

This library is based on the two following projects:

Dear ImGui : an amazing 'Immediate Mode GUI' C++ library

pyimgui : Python bindings for imgui (based on Cython).

imdebug, the image debugger :a neat image debugger / viewing by William Baxter has provided some ideas for the image_explorer feature.

Many thanks to their developers for their wonderful job.

Install & test:

Code::

git clone https://github.com/pthom/imgui_datascience.git cd imgui_datascience virtualenv env -p python3 source env/bin/activate pip install -r requirements.txt python run_example.py

Features

Display numpy.ndarray (aka opencv image)

The following types are supported : RGB, RGBA, GRAY, float32, float64

Code:

# returns mouse_position
imgui_cv.image(img, height=150, title="flowers")

Display matplotlib figures

Code:

figure = matplotlib.pyplot.figure()
x = numpy.arange(0.1, 100, 0.1)
y = numpy.sin(x) / x
plot.plot(x, y)

imgui_fig.fig(figure, height=250, title="f(x) = sin(x) / x")

Inspect images

• show pixels color (or float values)

• adjust visibility for float images

• save images

• zoom & pan (with possible sync between 2 images)

  

See https://www.youtube.com/watch?v=yKw7VaQNFCI&feature=youtu.be for an animated demo.

Code:

imgui_cv.image_explorer(img)

A simple way to run imgui programs

The simplest way to run a program a start adding gui buttons is shown below

Code:

def gui_loop():
    imgui.button("Click me")

def main():
    imgui_runner.run(gui_loop, imgui_runner.Params())

A simple way to quickly inspect images

Below is the simplest to quickly display any type of numpy array (RGB, float, etc) and to be able to inspect it.

Code:

image = ... # cv2.imread("...")
ImGuiImageLister.push_image("owl", image)
ImGuiLister_ShowStandalone()

Full demo

You can run a full demo using either

• Case 1 (from pip install):

Code:

pip install imgui_datascience
python -m imgui_datascience --example

• Case 2 (from checkout, with a virtualenv):

Code:

git clone https://github.com/pthom/imgui_datascience.git
cd imgui_datascience
virtualenv venv
source venv/bin/activate
pip install -r requirements.txt
pip install -r requirements_dev.txt
python run_example.py

• View the full demo (1'50") on youtube

click on the link below

https://www.youtube.com/watch?v=qstEZyLGsTQ&feature=youtu.be

Gotchas

Widget unique identifiers

Imgui identifies the widget through their label. If you have two buttons that have the same label, it might not differentiate them.

A workaround is to add "##" + an id after your label

Code:

if imgui.button("Click Me"):
    print("Clicked first button")
if imgui.button("Click Me##2"):
    print("Clicked second button")

Another workaround is to use imgui_ext.make_unique_label

Code:

if imgui.button(imgui_ext.make_unique_label("Click Me")):
    print("Clicked first button")
if imgui.button(imgui_ext.make_unique_label("Click Me")):
    print("Clicked second button")

OpenGL

This lib makes a heavy usage of OpenGL : it transfers the images from the RAM to you graphic card at each frame. The image textures are cached and only recreated if the image data has changed.

The library will detect that an image has changed by using a hash of its data. Two hash variant are possible :

• if imgui_cv.USE_FAST_HASH is set to True (which is default) : select 100 random pixels and hash them
• otherwise, compute the hash of the whole image data (using xxhash for performance)

You can change imgui_cv.USE_FAST_HASH value in order to change the behavior if needed.

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.