So i have been trying to detect and crop the face using mediapipe facemesh landmark points to calculate a bounding box .But when I ran the cropped image through the mediapipe it doesn’t detect the face at all for some images.
Can anybody explain why is it happening ?

I was under the impression that if the same face is detected by mediapipe that it should give the same detection results after cropping it using the same mediapipe model.

Thanks in advance