TreB1eN
"detect error" terminal output every time even face is available in the video feed
@TreB1eN
when I ran face_verify.py removed try catch to debug
got this error
traceback (most recent call last): File "face_verify.py", line 55, in <module> bboxes, faces = mtcnn.align_multi(image, conf.face_limit, conf.min_face_size)
anaconda3/envs/pytorch-face/lib/python3.7/site-packages/numpy/core/shape_base.py", line 234, in vstack return _nx.concatenate([atleast_2d(_m) for _m in tup], 0) ValueError: need at least one array to concatenate
but faces are available in the video its not detecting properly for face nearly 80 percent time face is not detecting 20 percent time its detecting when the face is not detecting i am getting the above error or this detect error
Is there any way to make face detection strong (or maybe I am doing someting wrong I have followed all the instructions but there were lot of errors so i have solved it may be due to that )
I have provide same video to opencv's haar cascade its detecting face when I ran face_verify.py its not even say unknow terminal out put is detect error
Yes I have saved it in facebank the error is it's not detecting face
Having the same problem, I think the MTCNN is not detecting faces as expected. even if there are no many images in the facebank, it should give 'UNKNOWN FACE'. @TreB1eN How can we improve the performance of the detection part (MTCNN)? Is there some kind of threshold, that we can tune. ?
@engmubarak48 thanks now its detecting face but what is the recognition algorithm its knn or something else
@engmubarak48 thanks now its detecting face but what is the recognition algorithm its knn or something else
@syedmustafa54 I also have the false positive issue, I think there is some threshold to tune. I will read the code and check.
About the recognition, please read Insightface repo. I don't think there is something related to KNN.
@syedmustafa54
There is 'threshold to decide identical faces' in face_verify.py script, can you try to tune this parameter. it might improve your performance.
@engmubarak48 thanks now its detecting face but what is the recognition algorithm its knn or something else
@syedmustafa54 I also have the false positive issue, I think there is some threshold to tune. I will read the code and check.
About the recognition, please read Insightface repo. I don't think there is something related to KNN.
solved it by controlling threshold
@engmubarak48 thanks now its detecting face but what is the recognition algorithm its knn or something else
@syedmustafa54 I also have the false positive issue, I think there is some threshold to tune. I will read the code and check. About the recognition, please read Insightface repo. I don't think there is something related to KNN.
solved it by controlling threshold
happy you solved your problem. which threshold value worked for you.?
@engmubarak48 in Insightface repo there is nothing about recognition code
@engmubarak48 thanks now its detecting face but what is the recognition algorithm its knn or something else
@syedmustafa54 I also have the false positive issue, I think there is some threshold to tune. I will read the code and check. About the recognition, please read Insightface repo. I don't think there is something related to KNN.
solved it by controlling threshold
happy you solved your problem. which threshold value worked for you.?
0.9 threshold value worked for me its better if we chat and share our knowledge here is my discord channel create for face https://discord.gg/WMkzTP lets discuss there
names = np.load(conf.facebank_path/'names.npy') there is a names.npy in this code , and what is it. If it is the idx list ,what is it format.
