Question

虽然我想在TensorFlowInferenceInterface中使用feed和fetch函数，但我无法理解feed和fetch args。

public void feed(String inputName, float[] src, long... dims) 
public void fetch(String outputName, float[] dst)

这是TensorflowInferenceInterface。↓ https://github.com/tensorflow/tensorflow/blob/r1.4/tensorflow/contrib/android/java/org/tensorflow/contrib/android/TensorFlowInferenceInterface.java

现在，我使用Android-Studio并希望使用MNIST导入程序这是制作协议缓冲区的程序。

import tensorflow as tf
import shutil
import os.path

if os.path.exists("./tmp/beginner-export"):
    shutil.rmtree("./tmp/beginner-export")

# Import data
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets("./tmp/data/", one_hot=True)

g = tf.Graph()

with g.as_default():
    # Create the model
    x = tf.placeholder("float", [None, 784])
    W = tf.Variable(tf.zeros([784, 10]), name="vaiable_W")
    b = tf.Variable(tf.zeros([10]), name="variable_b")
    y = tf.nn.softmax(tf.matmul(x, W) + b)

    # Define loss and optimizer
    y_ = tf.placeholder("float", [None, 10])
    cross_entropy = -tf.reduce_sum(y_ * tf.log(y))
    train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

    sess = tf.Session()

    # Train
    init = tf.initialize_all_variables()
    sess.run(init)

    for i in range(1000):
        batch_xs, batch_ys = mnist.train.next_batch(100)
        train_step.run({x: batch_xs, y_: batch_ys}, sess)

    # Test trained model
    correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))

    print(accuracy.eval({x: mnist.test.images, y_: mnist.test.labels}, sess))

# Store variable
_W = W.eval(sess)
_b = b.eval(sess)

sess.close()

# Create new graph for exporting
g_2 = tf.Graph()
with g_2.as_default():
    # Reconstruct graph
    x_2 = tf.placeholder("float", [None, 784], name="input")
    W_2 = tf.constant(_W, name="constant_W")
    b_2 = tf.constant(_b, name="constant_b")
    y_2 = tf.nn.softmax(tf.matmul(x_2, W_2) + b_2, name="output")

    sess_2 = tf.Session()

    init_2 = tf.initialize_all_variables();
    sess_2.run(init_2)

    graph_def = g_2.as_graph_def()

    tf.train.write_graph(graph_def, './tmp/beginner-export',
                         'beginner-graph.pb', as_text=False)

    # Test trained model
    y__2 = tf.placeholder("float", [None, 10])
    correct_prediction_2 = tf.equal(tf.argmax(y_2, 1), tf.argmax(y__2, 1))
    accuracy_2 = tf.reduce_mean(tf.cast(correct_prediction_2, "float"))
    print(accuracy_2.eval({x_2: mnist.test.images, y__2: mnist.test.labels}, sess_2))

输入的占位符名称为＆＃34;输入＆＃34;。
输出的占位符名称为＆＃34;输出＆＃34;。

请告诉我Feed并获取使用情况。

Answer 1

我已经提供了包含评论的示例代码。希望你能理解。

    private static final String INPUT_NODE = "input:0"; // input tensor name
    private static final String OUTPUT_NODE = "output:0"; // output tensor name
    private static final String[] OUTPUT_NODES = {"output:0"}; 
    private static final int OUTPUT_SIZE = 10; // number of classes
    private static final int INPUT_SIZE = 784; // size of the input
    INPUT_IMAGE //MNIST Image
    float[] result = new float[OUTPUT_SIZE]; // get the output probabilities for each class

    inferenceInterface.feed(INPUT_NODE, INPUT_IMAGE, 1, INPUT_SIZE); //1-D input (1,INPUT_SIZE)
    inferenceInterface.run(OUTPUT_NODES);
    inferenceInterface.fetch(OUTPUT_NODE, result);

对于我正在使用的Android Tensorflow库版本，我需要提供1-D输入。因此，Tensorflow代码需要根据那个进行修改，

x_2 = tf.placeholder("float", [None, 1, 784], name="input") //1-D input
x_2 = tf.reshape(x_2,[-1, 784]) // reshape according to the model requirements

希望这有帮助。

如何在TensorFlowInferenceInterface中使用feed和fetch函数？

1 个答案: