Known_face_locations In Face_encoding

[ojipadeson]

• face_recognition version: newest
• Python version: 3.7
• Operating System: windows 10

Description

In function

face_recognition.api.face_encodings(face_image, known_face_locations=None, num_jitters=1, model='small')

I used parameter known_face_locations, and the location is given by a SOTA face detection model.

bounding_box = [(left_top_x, left_top_y, right_bottom_x, right_bottom_y)]
face_encodings(image, known_face_locations = bounding_box)

Before I notice this parameter I did a manual face cut for every photo so that only face was in the image. Face recognition performed better after I did that. After I read the document, I found there should be another detection process in face_encoding. And I consider writing the bounding box to parameter known_face_locations will skip the built-in detection process and give me the same output. However, it performed even worse after I add this parameter. The output is unstable and always incorrect. What may be the problem?

Thanks for any feedback!