two questions,thanx

[dongdongmaster]

1.test.py,some of my jpgs can not detect feature,most is side face pics,but other algorithm works well,
2.what is the similarity value for face recognition as one person,0.6? 0.7?0.8?

[dongdongmaster]

for question1,face_embedding.py,"ret" is None,then returned

ret = self.detector.detect_face_limited(face_img, det_type = self.args.det)
if ret is None:
return None

[nttstar]

MTCNN failed, I think. You may need to change the parameters of MTCNN or use stronger facial landmarks detector to deal with it.

Cosine threshold may differ in different environments. To find a optimal value by cross validation.

[dongdongmaster]

thank u @nttstar

[dongdongmaster]

sorry,
"MTCNN failed, I think. You may need to change the parameters of MTCNN" ,where i can do this,change this value?
test.py
"parser.add_argument('--det', default=2, type=int, help='mtcnn option, 2 means using R+O, else using O')"

[nttstar]

thresholds, min-size, etc... Read the source code.

[brealisty]

for question1,face_embedding.py,"ret" is None,then returned

ret = self.detector.detect_face_limited(face_img, det_type = self.args.det)
if ret is None:
return None

I also meet the same problem，I do it in insight_pytorch get the face_bbox(ret not none), but this code is always return None. And I change the ret to 2 or some other interger, face_bbox performence not well