Harmonizing `mtcnn`and `tinyfaces`
The interfaces of bob.ip.facedetect.MTCNN
and bob.ip.facedetect.TinyFaces
differ.
MTCNN has two functions: detect
(returns list of bounding boxes, probabilities, landmarks) and annotations
(returns list of annotation dictionaries).
TinyFaces has only one function: detect
(returns list of annotation dictionaries), which is the equivalent of annotations
in the MTCNN case.
Finally, MTCNN provides a __call__
function that basically calls annotations
, which TinyFaces does not have such a function.
TODO:
- rename
TinyFaces.detect
toTinyfaces.annotations
- add
Tinyfaces.__call__
to callTinyfaces.annotations
- provide
TinyFaces.detect
similarly toMTCNN.detect
if this makes sense - Update documentation
- Clearly state in the documentation that eye landmarks are estimated, provide details where they are located, and show examples.