Face_landmark definition

[iainbryson]

We are currently defining it like this

trial_definition_specification = dict(trial_definition=dict(name='Landmarker calibration',
                                                            definition_data=dict(
                                                                    TrialType='Calibration',
                                                                    type='NewComponent::FaceLandmark',
                                                                    MaxNoOfAttempts='2',
                                                                    MinCalibrationAccuracyPct='20'
                                                                    )))

What we are thinking is to do it like this instead where we are including some extra parameters.

• this is the number of faces that we are expecting to track during the experiment. This is a standard parameter for the MediaPipe (it doesn’t have high priority) e.g. < num_faces >
• , this indicated if we want to save the Landmarks or not. The thinking is that we might want to reduce the computation and/or data transmission so only sending e.g. Blendshapes might be advantageous.
• < Blendshapes>, same as above <output_face_blendshapes >
• < CalibrationDuration>. See below. But essentially this would be a counter of how many seconds we want the participant to place their heads within the view of the camera.
• < FaceTransformation >. Indicate if the Facial transformation should be carried out or not. This is a standard mediapipe setting. <output_facial_transformation_matrixes>
• list of command separated indices
• don't send z coordinates

trial_definition_specification = dict(trial_definition=dict(name='Landmarker calibration',
                                                            definition_data=dict(
                                                                    TrialType='Calibration',
                                                                    type='FaceLandmark',
                                                                    NumberOfFaces = '2', # number of faces expected in the interface
                                                                    Landmarks=True, # return lanmark data
                                                                    Blendshapes=True, #  return blendshape
                                                                    FaceTransformation=True, # indcate if the affine transform should be performed or not
                                                                    CalibrationDuration='5' #duration of face within view measured in seconds
                                                                    )))

[iainbryson]

@madjens Regarding

FaceTransformation=True,  # indicate if the affine transform should be performed or not

This doesn't seem to match the configuration parameter for the landmarker:

[iainbryson]

Here's what's implemented now:

Property	Result
NumberOfFaces	Number of faces to detect.
Landmarks	Return Landmark data
BlendShapes	Return Blendshapes data
FaceTransformation	Return the face transformation matrices
StripZCoordinates	Strip Z coordinates from Landmarks data
IncludeBlandshapes	eyeLookInRight,eyeLookInLeft Comma separates list of Blendshapes to include. Removes the faceLandmarks not belonging the the included BlendShape

[iainbryson]

Also, there's potential to save a bunch of space with another option, say BlendshapeLandmarksOnly where we only send the landmarks for any of the (52) Blendshapes rather than the full 400+ points for the full face.

What do you think @madjens