slim_train_test.py 文件源码

def __setup_training(self,images, labels):
        tf.logging.set_verbosity(tf.logging.INFO)
        logits, end_points = self.network_fn(images)

        #############################
        # Specify the loss function #
        #############################
        loss_1 = None
        if 'AuxLogits' in end_points:
            loss_1 = tf.losses.softmax_cross_entropy(
                    logits=end_points['AuxLogits'], onehot_labels=labels,
                    label_smoothing=self.label_smoothing, weights=0.4, scope='aux_loss')
        total_loss = tf.losses.softmax_cross_entropy(
                logits=logits, onehot_labels=labels,
                label_smoothing=self.label_smoothing, weights=1.0)

        if loss_1 is not None:
            total_loss = total_loss + loss_1 



        global_step = slim.create_global_step()

        # Variables to train.
        variables_to_train = self.__get_variables_to_train()

        learning_rate = self.__configure_learning_rate(self.dataset.num_samples, global_step)
        optimizer = self.__configure_optimizer(learning_rate)


        train_op = slim.learning.create_train_op(total_loss, optimizer, variables_to_train=variables_to_train)

        self.__add_summaries(end_points, learning_rate, total_loss)

        ###########################
        # Kicks off the training. #
        ###########################

        slim.learning.train(
                train_op,
                logdir=self.train_dir,
                init_fn=self.__get_init_fn(),
                number_of_steps=self.max_number_of_steps,
                log_every_n_steps=self.log_every_n_steps,
                save_summaries_secs=self.save_summaries_secs,
                save_interval_secs=self.save_interval_secs)


        return