Face Recog-nition practices

[MyraBaba]

Hi,

I would like to know the best way to 1-to-many face recognition ways, and it could be help to the others. Accuracy increasing & clustering / searching methods

• As we know jittering image increasing the accuracy . From your experience is it useful to add more pictures of the same persons and use them to calculate one ID like jitter calculation (mean)? Or keep them separate and process jitter each of them and use them seperately. ? What could be the other methods to increase accuracy and decrease false alarms?

• We have two different face database : 1 - 10.000 people and 2 - 100.000 . All ID extracted with jittering image. To search 1-to-many what is the best way :

  • check all lenght(face_ID_1 - faceID_2) walk through all faces and take top few which least distance.
  • Or make a chineese clustering and ask him to make cluster for searched face.

  *any other suggestion..

Best

thanks again davis such a great library. you have lifetime ticket from us for beer/wine/lunch :) if you pass by .. as a friend.

[davisking]

Don't average a bunch of vectors from different images. A better thing to do is to use either k-nearest-neighbors or something like a linear SVM if you want to take advantage of multiple images. I wouldn't use chinese whispers.

[MyraBaba]

If we have only one image per person that the is the best option to check lenght(face_ID_1 - faceID_2) and find the (with sort) smallest distance ?

[davisking]

Generally speaking, yes.

[ksachdeva]

Hi Davis,

Don't average a bunch of vectors from different images. A better thing to do is to use either k-nearest-neighbors or something like a linear SVM if you want to take advantage of multiple images. I wouldn't use chinese whispers.

Would appreciate if you could provide guidance/education on why Chinese Whisper should not be used.

I have a similar challenge i.e. doing 1-to-many match.

Regards & thanks Kapil

[wyd1520]

@MyraBaba 将所有人人脸特征截入到内存里，然后再用 face_ID_1=xxx; foreach item in cacheFeatures lenght(face_ID_1 - item) 花不了几毫秒

[dlib-issue-bot]

Warning: this issue has been inactive for 28 days and will be automatically closed on 2018-09-13 if there is no further activity.

If you are waiting for a response but haven't received one it's likely your question is somehow inappropriate. E.g. you didn't follow the issue submission instructions, or your question is easily answerable by reading the FAQ, dlib's documentation, or a Google search.

[dlib-issue-bot]

Warning: this issue has been inactive for 35 days and will be automatically closed on 2018-09-13 if there is no further activity.

If you are waiting for a response but haven't received one it's likely your question is somehow inappropriate. E.g. you didn't follow the issue submission instructions, or your question is easily answerable by reading the FAQ, dlib's documentation, or a Google search.

[dlib-issue-bot]

Notice: this issue has been closed because it has been inactive for 37 days. You may reopen this issue if it has been closed in error.