Align faces when only bounding boxes are provided
The BoundingBoxAnnotatorCrop
cropper uses a 2 stage strategy to crop and align faces in case annotation_type
has a bounding-box.
In the first stage, it crops the face using the {topleft
, bottomright
} parameters and expands them using a margin
factor.
In the second stage, it uses the annotator
to estimate {leye
and reye
} to make the crop using :py:class:bob.ip.base.FaceEyesNorm
.
In case the annotator doesn't work, it returns the cropped face using the bounding-box
coordinates.
It's being a while that I wanted to put this in place.
Follow the impact of this on the test1
IJB-C protocol (old 1:1
) and on the test4
protocols (open-set identification protocols)
test1.pdf dir-test4-g1.pdf dir-test4-g2.pdf
Do you have sometime to review this one @lcolbois ?
Thanks