Question

我保存了训练模型。

预测代码如下：

get_serializer

我使用星图多进程来获取预测值：

def predict(predict_filename):
    with tf.Session() as sess:

        # normalization image
         ...
        im_norm = tf.image.per_image_standardization(rgb_image)

        sess = tf.InteractiveSession()
        with sess.as_default():

            coord = tf.train.Coordinator()
            threads = tf.train.start_queue_runners(coord=coord, sess=sess)

            def get_ckpt():
                ckpt = tf.train.get_checkpoint_state(FLAGS.checkpoint_dir)
                return ckpt

            ckpt = get_ckpt()

            def get_saver():
                new_saver = tf.train.import_meta_graph(ckpt.model_checkpoint_path + '.meta')
                return new_saver

            new_saver = get_saver()

            if ckpt and ckpt.model_checkpoint_path:
                print(ckpt.model_checkpoint_path)
                new_saver.restore(sess, ckpt.model_checkpoint_path)
                try:
                    while not coord.should_stop():

                         result = sess.run("Linear_regression/Add:0", feed_dict={'conv1/Reshape:0': image_new.eval(),                                                                         'dropout2/dropout/keep_prob:0': 0.2})
                        ...
                except tf.errors.OutOfRangeError:
                    print('Done training after reading all data')
                finally:  # When done, ask the threads to stop.
                    coord.request_stop()
                    # Wait for threads to finish.
                    # tf.reset_default_graph()
                    coord.join(threads)
                    sess.close()
                    # os._exit(0)
            else:
                print('no checkpoint found')

这里我只使用了一个过程。运行项目，＆＃34; 1.jpg＆＃34;的预测值。可以成功，然后得到错误：

 def main(argv=None):  
        filename = []
        filename.append('./1.jpg')
        filename.append('./2.jpg')
        filename.append('./3.jpg')
        filename = np.array(filename).astype(str)
        print(filename)
        pool = ThreadPool(1)
        pool.starmap(predict, zip(filename))
        pool.join()
        pool.close()

我真的不知道为什么第二张图片的预测得到错误，但第一张图片能得到正确的结果？

当我在＆＃34;最后＆＃34;中添加os._exit（0）时，一切顺利，三张图片都可以得到正确的结果，为什么？

如果添加，如果我添加sess.run（tf.global_variables_initializer（）），在预测第一张图片后，再获取日志：Traceback（最近一次调用最后一次）：

tensorflow.python.framework.errors_impl.FailedPreconditionError: Attempting to use uninitialized value file_queue/limit_epochs/epochs
     [[Node: file_queue/limit_epochs/CountUpTo = CountUpTo[T=DT_INT64, _class=["loc:@file_queue/limit_epochs/epochs"], limit=40, _device="/job:localhost/replica:0/task:0/cpu:0"](file_queue/limit_epochs/epochs)]]

如何通过多进程预测恢复模型

0 个答案: