pupil-detectors 2.0.1

Pupil detectors

pupil-detectors

This Python package contains the standalone 2D pupil detectors for the Pupil-Labs software stack.

Installation

macOS and Linux

This package has a couple of non-python dependencies that you will have to install yourself before you can install pupil-detectors. We are working on clean setup instructions, but until then you can just install all dependencies from the Pupil software stack. You can find installation guides on the Pupil GitHub page (section: Installing Dependencies).

Then you can install pupil-detectors with pip:

pip install pupil-detectors

Windows

Since the dependency setup is very complex on Windows, we provide prebuilt-wheels. This means you don't have to install any dependencies and can just install pupil-detectors:

pip install pupil-detectors

If you don't want to use our prebuilt versions, see the section Building from Source further below.

Usage

Here's a quick example on how to detect and draw an ellipse.

import cv2
from pupil_detectors import Detector2D

detector = Detector2D()

# read image as numpy array from somewhere, e.g. here from a file
img = cv2.imread("pupil.png")
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

result = detector.detect(gray)
ellipse = result["ellipse"]

# draw the ellipse outline onto the input image
# note that cv2.ellipse() cannot deal with float values
# also it expects the axes to be semi-axes (half the size)
cv2.ellipse(
    img,
    tuple(int(v) for v in ellipse["center"]),
    tuple(int(v / 2) for v in ellipse["axes"]),
    ellipse["angle"],
    0, 360, # start/end angle for drawing
    (0, 0, 255) # color (BGR): red
)
cv2.imshow("Image", img)
cv2.waitKey(0)

Developers

Building from Source

You can install this package locally from source. Make sure you have all necessary dependencies setup (see Installation section above).

NOTE: When using a Python virtual environment, it should be created without the --system-site-packages flag; otherwise building might fail.

NOTE: For Windows the dependency setup is quite complex. Until we have clean instructions, please follow the guide for setting up Windows dependencies for Pupil.

NOTE: When building the package on your own, you can experience severe performance differences when not having setup your dependencies correctly. Make sure to compare performance to prebuilt wheels or to bundled versions of Pupil.

# Clone repository
git clone [e-mail]:pupil-labs/pupil-detectors.git
cd pupil-detectors

# Install from source with development extras
pip install ".[dev]"

# Run tests
pytest tests

Maintainers

Distribution

This project does not currently support automated distribution. Steps for a new release:

1. Make sure the code works.
2. Install dependencies

pip install bump2version twine

3. Create new tag with bump2version.

   • Decide on type of version bump: major/minor/patch

   • Make sure the working directory is clean! (git status)

   • Modify CHANGELOG.md to include newest version notes.

   • Stage changelog (don't commit):

     git add CHANGELOG.md

   • Run the appropriate bump2version. This will create a new commit with all necessary changes (including the staged changelog) and a new tag!

     # ONLY ONE OF THOSE!
     bump2version major --allow-dirty
     # or
     bump2version minor --allow-dirty
     # or
     bump2version patch --allow-dirty
     

4. Push the new commit and (all) tags.

git push
git push --tags

5. Build the source distribution and wheel (Windows). Use the internal bundle-machine for Pupil for the correct dependency setup!

python setup.py sdist
pip wheel --no-deps . -w dist

6. Test installing from wheel and from sdist!

7. Upload wheel and sdist to PyPI.

twine upload ./dist/*

Changelog

2.0.1 (2020-12-03)

Removed all remaining Ceres references. Eigen3 remains a dependency.

2.0.0 (2020-12-02)

Removed Detector3D - #17

In favor of our new pye3d detector, we have removed the previous Detector3D class. This allows us to remove Ceres as a dependency and to cleanup the detector class interface.

Specifically, getting and setting properties has been simplified by removing property namespaces.

- NamespacedProperties = T.Dict[str, T.Dict[str, T.Any]]
+ DetectorProperties = T.Dict[str, T.Any]

- def get_property_namespaces(self) -> T.Iterable[str]

- def get_properties(self) -> NamespacedProperties:
+ def get_properties(self) -> DetectorProperties:

- def update_properties(self, properties: NamespacedProperties) -> None
+ def update_properties(self, properties: DetectorProperties) -> None

The Roi class has been moved from pupil_detectors.utils to pupil_detectors.roi.

1.1.1 (2020-08-26)

Improvements

• Added default lookup paths for OpenCV on Ubuntu 20.04, enabling building from source out of the box there.
• Added parameter to specify the focal length of the camera supplying the eye images for a more accurate 3D model.

1.1.0 (2020-05-04)

Changed

• Changed the default 2D detector properties to be the same as the default overrides that the 3D detector applies.

1.0.5 (2020-04-20)

Added

• Added option to run 3D detector without internal 2D detector, but from serialized data of external 2D detector

1.0.4 (2020-01-13)

Fixed

• Fixed crash when installing from source distribution package

1.0.3 (2020-01-07)

Fixed

• Wrong Roi.rect() computation.

1.0.2 (2019-12-03)

• Initial release.