EyepadAlign for aligning face images across different datasets

[vmirly]

Description

Adding a new module called eyepad_align that aligns face images by fitting the location of eyes.

Furthermore, this technique can be used for aligning images across multiple datasets, where the images in one dataset will be transformed to the average location of eyes in another dataset.

Related issues or pull requests

Pull Request Checklist

• Added a note about the modification or contribution to the ./docs/sources/CHANGELOG.md file (if applicable)
• Added appropriate unit test functions in the ./mlxtend/*/tests directories (if applicable)
• Modify documentation in the corresponding Jupyter Notebook under mlxtend/docs/sources/ (if applicable)
• Ran nosetests ./mlxtend -sv and make sure that all unit tests pass (for small modifications, it might be sufficient to only run the specific test file, e.g., nosetests ./mlxtend/classifier/tests/test_stacking_cv_classifier.py -sv)
• Checked for style issues by running flake8 ./mlxtend

[pep8speaks]

Hello @vmirly! Thanks for updating the PR.

• There are no PEP8 issues in the file mlxtend/externals/pyprind/__init__.py !

• In the file mlxtend/externals/pyprind/generator_factory.py, following are the PEP8 issues :

Line 23:1: E305 expected 2 blank lines after class or function definition, found 1

• There are no PEP8 issues in the file mlxtend/externals/pyprind/prog_class.py !

• In the file mlxtend/externals/pyprind/progbar.py, following are the PEP8 issues :

Line 75:80: E501 line too long (85 > 79 characters)
Line 76:48: E128 continuation line under-indented for visual indent

• There are no PEP8 issues in the file mlxtend/externals/pyprind/progpercent.py !

• There are no PEP8 issues in the file mlxtend/image/__init__.py !

• There are no PEP8 issues in the file mlxtend/image/extract_face_landmarks.py !

• In the file mlxtend/image/eyepad_align.py, following are the PEP8 issues :

Line 153:80: E501 line too long (81 > 79 characters)

• There are no PEP8 issues in the file mlxtend/image/tests/test_extract_face_landmarks.py !

• There are no PEP8 issues in the file mlxtend/image/tests/test_eyepad_align.py !

• There are no PEP8 issues in the file mlxtend/image/utils.py !

Comment last updated on December 09, 2018 at 18:35 Hours UTC

[coveralls]

Coverage decreased (-0.2%) to 91.619% when pulling ac80d96 on vmirly:vm-eyepad into 18f0080 on rasbt:master.

[rasbt]

nice idea to use scikit-learn-like API for this. This way, it can be technically also used as a transformer in a scikit-learn pipeline.

[rasbt]

Adding a new class, mlxtend.image.EyepadAlign, that aligns face images by based on the eye location. Furthermore, this technique can be used for aligning images across multiple datasets, where the images in one dataset will be transformed to the average location of eyes in another dataset.

I think this would be a nice summary for the CHANGELOG then as well

[rasbt]

I think we should add a warning that individual pixel positions can be slightly different by 1-2 pixels on windows if users run the code on windows.

I can do that

[rasbt]

List of the todos we discussed as reference, so we don't forget

• maintaining the folder structure
• add option for nose-center alignment
• add documentation to all methods
• extend unit tests