1. A trained person high accuracy
2. Is to a certain degree of model training for everyone still new to one training once again?
Big companies such as pay treasure to pay brush face is how to solve this problem, when they enter input only once, but the accuracy is high, is that they already have our face library?
That if it is a small company to do the little people face recognition algorithm, how to add a new person is automatically retraining and automatic replacement model? Because every trained model of face images to extract feature, it's trouble

CodePudding user response:

Face recognition problem generally divided into the following three:
1. The face detection
2. The facial feature points positioning and correct
3. Contrast/face recognition

1 and 2 points, as if it is not you attention to skip the first

Whether large companies or small companies, will not add one training model,
Face contrast or identification of the model, the input is an image, the output is a feature vector (note that is not a specific person
Through the contrast has some characteristics of vector to determine whether the model library, and who is

Now the problem is how to train a model, the features of the output vector to be possible to distinguish between different people, probably knew the twin network, three yuan loss, such as interested to see more

CodePudding user response:

reference 1st floor Zhu Mingde response:

face recognition problem generally divided into the following three:
1. The face detection
2. The facial feature points positioning and correct
3. Contrast/face recognition

1 and 2 points, as if it is not you attention to skip the first

Whether large companies or small companies, will not add one training model,
Face contrast or identification of the model, the input is an image, the output is a feature vector (note that is not a specific person
Through the contrast has some characteristics of vector to determine whether the model library, and who is

Now the problem is how to train a model, the features of the output vector to be possible to distinguish between different people, probably knew the twin network, three yuan, such as the loss are interested to see more

Yes, the problem is how to train model to the problem of the model applicable for all objects, I know, three yuan loss is not big amount of data to use triple facenet - loss? To twenty people replace my training center - loss, the training is to a certain degree to a great amount of data (including your own face, the rest with public face database available?) To make it more than 99% accuracy

CodePudding user response:

In the process of training is need a large amount of data (different face) to the training, so can make model can learn how to extract different facial features, when after new join a human face, the trained model can extract to the right the right characteristics, formed a characteristic vectors stored in the local, when testing, with the characteristics of the real-time characteristic vectors generated and saved locally vector similarity matching,

CodePudding user response:

Whether it's 1:1 ratio of a face, or 1: N, M: N process of face recognition are similar,

CodePudding user response:

1: N, M: N process of face recognition are similar,