使用Java API实现维度0的切片索引0越界

Question

import base64
import numpy as np
import tensorflow as tf
​
​
fn_load_image = lambda filename: np.array([base64.urlsafe_b64encode(open(filename, "rb").read())])
filename='test.jpg'
with tf.Session() as sess:
    loaded = tf.saved_model.loader.load(sess, ['serve'], 'tools/base64_model/1')
    image = fn_load_image(filename)
    p = sess.run('predictions:0', feed_dict={"input:0": image})
    print(p)

    // load the model Bundle
    try (SavedModelBundle b = SavedModelBundle.load("tools/base64_model/1",
            "serve")) {

        // create the session from the Bundle
        Session sess = b.session();

        // base64 representation of JPG
        byte[] content = IOUtils.toByteArray(new FileInputStream(new File((args[0]))));

        String encodedString = Base64.getUrlEncoder().encodeToString(content);

        Tensor t = Tensors.create(encodedString);

        // run the model and get the classification
        final List<Tensor<?>> result = sess.runner().feed("input", 0, t).fetch("predictions", 0).run();

        // print out the result.
        System.out.println(result);
    }

这应该是等价的，即我发送一个图像的base64表示给一个模型，我得到了一个异常

线程“main”java.lang.IllegalArgumentException中的异常：维度0的切片索引0越界。[[{{node map/strided_slice}}]]在org.tensorflow.session.run（本机方法）在org.tensorflow.session.access$100(Session.java:48)在org.tensorflow.session$runner.runhelper(Session.java:326)在org.tensorflow.session$runner.java:276)，在com.stolencamerafinder.storm.crawler.bolt.encryments.hellotensorflow.main（

张量应该有不同的内容吗？以下是saved_model_cli告诉我的模型。

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['serving_default']:
  The given SavedModel SignatureDef contains the following input(s):
    inputs['inputs'] tensor_info:
        dtype: DT_STRING
        shape: (-1)
        name: input:0
  The given SavedModel SignatureDef contains the following output(s):
    outputs['outputs'] tensor_info:
        dtype: DT_FLOAT
        shape: (-1, 2)
        name: predictions:0
  Method name is: tensorflow/serving/predict

Answer 1

您的模型期望输入张量为1级，而您提供的张量为0级。

这一行产生一个可变长度的标量张量（即dt_string)。

Tensor t = Tensors.create(encodedString);

但是，预期的张量是秩为1的，从这里的形状(-1)中可以看到，这意味着它预期的是一个由不同数量的元素组成的向量。

The given SavedModel SignatureDef contains the following input(s):
    inputs['inputs'] tensor_info:
        dtype: DT_STRING
        shape: (-1)
        name: input:0

// base64 representation of JPG
byte[] content = IOUtils.toByteArray(new FileInputStream(new File((args[0]))));
byte[] encodedBytes = Base64.getUrlEncoder().encode(content);
Tensor t = Tensors.create(new byte[][]{ encodedBytes });
...