Confidence <.8

[sorora-b]

📕 The doc issue

I refer to this part of the documentation:
"Along with everything else, OpenFace outputs ‘confidence scores’ that delineate how confident it is that it is indeed seeing a face in an image from a video. If the confidence score for an image frame is below 80%, we do not process any facial activity variables for those frames and the framewise raw variable output does not contain values. All derived facial variables that are then calculated from the raw variables only reflect image frames where the confidence was >80%. " on page 23.

In the raw output variable output I can see some confidences <.8 and the AU-data does contain values. I guess these frames are not included into the calculation of the derived variables but is there a possibility to see which frames were dismissed or can I be sure that all the frames with a confidence <.8 were dismissed?

Suggest a potential alternative/fix

No response

[ADParedes]

Hi, and great question. This would require functionality to be built to support this use case. Right now OpenDBM integrates the out-of-the-box functionality of Openface which is from where we are inheriting this limitation. However, we are tracking enhancement requests. If you could provide a little more information on the user story and its impact on your work, I can open up a future enhancement request to the open-source community.