Face Recognition issue with mobile facenet

Question

I am facing the following issue while using a Mobile FaceNet for face recognition:

1. Sometimes it identifies the wrong face. For example, I add the face of A person and B person

On scanning person A face it identifies it as person B's face

2. sometimes it identifies randomly any other person who is not even registered in the database

NOTE: the face embedding is saved in the MySQL database as GSON text.

Kindly suggest some strong accurate face recognition library that does not has such kind of bugs

Or guide me on how to get rid of these bugs

    private final TensorImage loadImage(Bitmap bitmap) {
        TensorImage var10000 = this.tImage;

        var10000.load(bitmap);
        ImageProcessor imageProcessor = (new ImageProcessor.Builder()).add((ImageOperator) (new ResizeOp(this.inputImageHeight, this.inputImageWidth, ResizeOp.ResizeMethod.BILINEAR))).add((TensorOperator) (new NormalizeOp(this.IMAGE_MEAN, this.IMAGE_STD))).build();
        TensorImage var10001 = this.tImage;


        Object var3 = imageProcessor.process(var10001);

        return (TensorImage) var3;
    }

  private final float[] getEmbedding(Bitmap bitmap) {
        this.tImage = this.loadImage(bitmap);
        Interpreter var10000 = this.interpreter;

        TensorImage var10001 = this.tImage;

        ByteBuffer var4 = var10001.getBuffer();
        TensorBuffer var10002 = this.tBuffer;

        var10000.run(var4, var10002.getBuffer().rewind());
        TensorBuffer var2 = this.tBuffer;


        float[] var3 = var2.getFloatArray();

        return var3;
    }

    private final float cosineSimilarity(float[] A, float[] B) {
        if (A != null && B != null && A.length != 0 && B.length != 0 && A.length == B.length) {
            double sumProduct = 0.0D;
            double sumASq = 0.0D;
            double sumBSq = 0.0D;
            int i = 0;

            for (int var10 = A.length; i < var10; ++i) {
                sumProduct += (double) (A[i] * B[i]);
                sumASq += (double) (A[i] * A[i]);
                sumBSq += (double) (B[i] * B[i]);
            }

            return sumASq == 0.0D && sumBSq == 0.0D ? 2.0F : (float) (sumProduct / (Math.sqrt(sumASq) * Math.sqrt(sumBSq)));
        } else {
            return 2.0F;
        }
    }

here I have attached code that is used on the android side for getting embedding and matching faces.