Triplet Loss

triplet_loss.py

def triplet_loss(_, y_pred, triplet_margin=1):
    '''
        Assume: y_pred shape is (batch_size, 2)
        
        Example for how to construct such a corresponding triplet network:
        
        def euclidean_distance(vects):
            x, y = vects
            return K.sqrt(K.maximum(K.sum(K.square(x - y), axis=1, keepdims=True), K.epsilon()))
            
        def trip_output_shape(shapes):
            shape1, shape2 = shapes
            return (shape1[0], 1)
            
        def build_triplet(embed_network, input_shape):
            img_ref  = Input(shape=input_shape)
            img_pos  = Input(shape=input_shape)
            img_neg  = Input(shape=input_shape)
            embed_ref = embed_network(img_ref)
            embed_pos = embed_network(img_pos)
            embed_neg = embed_network(img_neg)

            distance_pos = Lambda(euclidean_distance, output_shape=dist_output_shape)([embed_ref, embed_pos])
            distance_neg = Lambda(euclidean_distance, output_shape=dist_output_shape)([embed_ref, embed_neg])

            output_layer = Lambda(lambda vects: K.concatenate(vects, axis=1))([distance_pos, distance_neg])

            model =    Model(  inputs=[img_ref, img_pos, img_neg],
                                outputs = output_layer,
                                name='triplet-network')
            return model
    '''
    margin = K.constant(triplet_margin)
    subtraction = K.constant([1, -1], shape=(2, 1))
    diff =  K.dot(K.square(y_pred), subtraction)

    loss = K.maximum(K.constant(0), margin + diff)

    return loss