CODEWITHSUNDEEP
tensorflow
Peter Mourfield
Peter Mourfield

Reputation: 1885

Using Tensorflow saved_model, getting ValueError: Cannot feed value of shape () for Tensor 'Placeholder_1084:0', which has shape '(?,)'

I'm using Tensorflow 1.14.0 and am trying to write a Python script that loads a model saved using tf.compat.v1.saved_model.simple_save to process an image from disk. The model works fine when I use Tensorflow Serving and a GRPC or REST client. However, when I try to use it in a single script, I get the following error:

ValueError: Cannot feed value of shape () for Tensor 'Placeholder_1084:0', which has shape '(?,)'

I'm sure that I've got something fundamental wrong but haven't been able to figure it out just yet. I've put together a short sample that illustrates what I'm trying to accomplish:

#!/usr/bin/env python

import tensorflow as tf
from tensorflow.python.saved_model import tag_constants
from tensorflow.python.saved_model import signature_constants

savedModelPath = "./models/inception_v3/1/"
filename = "./imagenet/n02763714/image_0.jpg"

session = tf.compat.v1.Session(graph=tf.Graph())
with session.graph.as_default():
    model = tf.compat.v1.saved_model.load(export_dir=savedModelPath, sess=session, tags=[tag_constants.SERVING])

    model_def = model.signature_def[signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY]

    graph_def = tf.compat.v1.get_default_graph()
    input_tensor = graph_def.get_tensor_by_name(model_def.inputs["image_bytes"].name)
    output_tensor = graph_def.get_tensor_by_name(model_def.outputs["predictions"].name)

    image_data = tf.io.gfile.GFile(filename, 'rb').read()

    session.run(output_tensor, {input_tensor: image_data})

Thanks in advance for any help!

EDIT: If it's helpful here is the model signature:

inputs {
  key: "image_bytes"
  value {
    name: "Placeholder_1084:0"
    dtype: DT_STRING
    tensor_shape {
      dim {
        size: -1
      }
    }
  }
}
outputs {
  key: "predictions"
  value {
    name: "inception_v3/predictions/Softmax:0"
    dtype: DT_FLOAT
    tensor_shape {
      dim {
        size: -1
      }
      dim {
        size: 1000
      }
    }
  }
}
method_name: "tensorflow/serving/predict"

Upvotes: 1

Views: 95

Answers (1)

javidcf
javidcf

Reputation: 59691

Your model expects a one-dimensional input, which is what the shape (?,) means. Presumably, in this input each value is a sequence of bytes encoding an image. In your code, you seem to be trying to use the model to get the prediction for a single image. When you do:

image_data = tf.io.gfile.GFile(filename, 'rb').read()

You are getting a single bytes value in image_data, which, from the point of view of TensorFlow, is a scalar value (a value with shape ()), like an individual int or float. You can get it to work if you give that value as a one-element list with that value, which TensorFlow will interpret as a one-dimensional input:

session.run(output_tensor, {input_tensor: [image_data]})

Note that the value returned by the line above will (again, presumably) have shape (1, 1000). The predicted output vector for your image can then be extracted as:

session.run(output_tensor, {input_tensor: [image_data]})[0]

Upvotes: 1

Related Questions