different croped face from draw_detections and extract Api for same image

[Suma3]

Hi..
I was doing testing on the diiferent api's of the repo & realised one thing that there is huge differnece between bouding box that has been drwan around face using /draw_detections and face data written from /extract api while setting return_face_data = True for the same image.I am getting very low performace on images so I started dig down in the code and i realised that while getting embedding from model you are not using bounding box area instead you are making affine transformation around landmarks so any particular reason for this why not bounding box ?

for reference i have added images you can see there is extra pixels in last image compare to first image

[SthPhoenix]

This is default preprocessing required for insightface recognition models.
Face should be aligned and centered based on landmarks for maximum recognition accuracy.
Bounding box is used mostly for display purposes or might be used in conjunction with separate landmarks detector

[SthPhoenix]

Closing for inactivity.