Question

我正在尝试实现三重损失功能以进行人脸识别。任何人都可以澄清以下内容：

处理完每张图像后，我使用dlib库获得128-float64值（编码器）。
我已经对人（图像，编码器）（数十万人）的数据库进行了预配对

接下来我创建一个网络：

def create_base_net_simple(input):
    x = Flatten()(input)
    x = Dense(128, activation='relu')(x)
    x = Dropout(0.1)(x)
    x = Dense(128, activation='relu')(x)
    x = Dropout(0.1)(x)
    x = Dense(128, activation='relu')(x)
    return Model(input, x)

def triplet_merge(inputs):
    a, p, n = inputs
    return K.sum(a * (p - n), axis=1)

def triplet_merge_shape(input_shapes):
    return (input_shapes[0][0], 1)

def triplet_loss(y_true, y_pred):
    return -K.mean(K.log(K.sigmoid(y_pred)))

def build_model(n_in):
    a = Input(shape=(n_in,))
    p = Input(shape=(n_in,))
    n = Input(shape=(n_in,))

    base_model = create_base_net_simple(Input(shape=(n_in, 1)))
    a_model = base_model(a)  # anchor
    p_model = base_model(p)  # pos
    n_model = base_model(n)  # neg

    e = merge([a_model, p_model, n_model], mode=triplet_merge, output_shape=triplet_merge_shape)
    model = Model(input=[a, p, n], output=e)
    model.compile(loss=triplet_loss, optimizer='rmsprop')

    return model

并使用我的数据库对其进行训练：

anch = [...]  # array of anchor Encoders (128-vector)
pos = [...]  # array of positive Encoders - same person
neg = [...]  # array of neg Encoders - random other person
# so anch[i] is encoder of same person as pos[i], but neg[i] is different person
labels = [...]  # array of zeros
n_in = 128  # encoder len
model = build_model(n_in)
model.fit([anch, pos, neg], labels, batch_size=2048, nb_epoch=60)
net_name = cur_dir + os.sep + 'TrainedModel.h5'
model.save_weights(net_name)

然后-如果两个随机编码器是同一个人或不同，我该如何预测？像这样：
```
similarity_lvl = model_predict.predict([enc0, enc1]) # enc0, enc1 - encoders of random persons
```
我是否必须创建另一个模型并将受过训练的权重加载到其中？（如果是，应该是哪种模型？）

使用python编码器的三重态损耗预测

0 个答案: