Matching names to faces. about face-api.js

[justadudewhohacks]

Hi,

There are no trained models saved for the face recognition part. Basically, in the face recognition examples is an example image loaded for each character and the face descriptor is computed and stored in an array (reference descriptors). Note, that the images loaded here are already aligned face images, thus you can pass them straight into the FaceRecognitionNet.

Now, in the example detectAndRecognizeFaces, once you select an input image, all faces are detected in that image and their face descriptors are computed. What we do now is to match these face descriptors (query descriptors) to the reference descriptors, by computing the euclidean distance between two descriptors and simply find the one with the lowest distance, to identify the person with the help of the reference data.

I would recommend you to also read my introduction, where we actually do the same thing: https://itnext.io/face-api-js-javascript-api-for-face-recognition-in-the-browser-with-tensorflow-js-bcc2a6c4cf07.

One thing that might be confusing in the example is, that you can have multiple images / descriptors for each class in your reference data instead of only 1 descriptor, which should make it more robust. Now, whats done in getBestMatch, is to compute the distance of the query descriptor to all reference descriptors for a class and simply return the mean value of the distances.

Hope that clarifies it.