Skip to content

Latest commit

 

History

History
282 lines (190 loc) · 8.81 KB

DOCUMENTATION.md

File metadata and controls

282 lines (190 loc) · 8.81 KB
Raw

TensorFlow.js Graph Converter API

Modules

The converter comes with several Python modules:

tfjs_graph_converter.api

Functions to load TensorFlow graphs from TensorFlow.js graph models and convert TensorFlow.js graph models to TensorFlow frozen graph and SavedModel formats.

Functions

tfjs_graph_converter.api.load_graph_model

Load a tensorflowjs graph model from a directory and return a tensorflow (v1) graph.

Arguments

  • model_dir: Name of the directory that contains the model.json and the weights

Returns

tensorflow Graph object (frozen graph) that can be used for inference or converted into different formats

Example

import numpy as np
import tensorflow as tf
import tfjs_graph_converter as tfjs

# load TFJS model into tensorflow frozen graph
graph = tfjs.api.load_graph_model('~/some-website/saved_model_stylize_js/')

# load image into numpy array
img = tf.keras.preprocessing.image.load_img('~/images/sample.jpg')
x = tf.keras.preprocessing.image.img_to_array(img, dtype=np.float32)
# scale input to [0..1] and expand dimensions to 4d array
x /= 255.0
content_image = x[tf.newaxis, ...]

# evaluate the loaded model directly
with tf.compat.v1.Session(graph=graph) as sess:
    # the module provides some helpers for querying model properties
    input_tensor_names = tfjs.util.get_input_tensors(graph)
    output_tensor_names = tfjs.util.get_output_tensors(graph)
    input_tensor = graph.get_tensor_by_name(input_tensor_names[0])

    results = sess.run(output_tensor_names, feed_dict={input_tensor: content_image})

# save result
stylised = np.squeeze(results)
tf.keras.preprocessing.image.save_img('~/images/stylized_sample.jpg', stylised)

tfjs_graph_converter.api.graph_model_to_frozen_graph

Convert a TensorFlow.js graph model to a tensorflow frozen graph.

Arguments

  • model_dir: Name of the directory that contains the model.json and the weights
  • export_path: Path to the frozen graph file

Example

import tfjs_graph_converter as tfjs

# convert TFJS model to a frozen graph
tfjs.api.graph_model_to_frozen_graph(
    '~/some-website/saved_model_stylelize_js/', '~/models/stylize.pb')

tfjs_graph_converter.api.graph_model_to_saved_model

Convert a TensorFlow.js graph model to a tensorflow SavedModel on disk.

Arguments

  • model_dir: Name of the directory that contains the model.json and the weights
  • export_dir: Directory to store the tensorflow SavedModel in
  • tags: Array of strings that are annotations to identify the graph (e.g. ['serve'])

Example

import tfjs_graph_converter as tfjs

# convert TFJS model to a SavedModel
tfjs.api.graph_model_to_saved_model('~/some-website/saved_model_stylelize_js/',
'~/models/stylize/', ['serve'])

tfjs_graph_converter.api.graph_models_to_saved_model

Merge several TensorFlow.js graph models into a single SavedModel. Separate models are identified by different tags (see documentation)

Arguments

  • model_list: List of tuples containing the TensorFlow.js graph model directory and a list of tags for the imported model. Format example: [('/path/to/1st/model_json/', ['serve', 'preprocess']), ('/path/to/2nd/model_json/', ['serve', 'predict'])]
  • export_dir: Directory to store the combined SavedModel in

Example

import tfjs_graph_converter as tfjs

model_list = [
    ('~/website/preprocess_saved_model_js/', ['serve', 'preprocess']),
    ('~/website/predict_saved_model_js/', ['serve', 'predict']),
    ('~/website/finalize_saved_model_js/', ['serve', 'finalize'])
]
# convert TFJS model to a SavedModel
tfjs.api.graph_models_to_saved_model(model_list, '~/models/combined/')

tfjs_graph_converter.util

Contains utility functions to query graph properties such as input- and output tensor names.

Contents

tfjs_graph_converter.util.NodeInfo

namedtuple with the following fields:

  • name: Name of the graph node (e.g. layer name)
  • shape: Shape associated with the tensor (input or output) as a list of sizes (e.g. [1, 256, 256, 3])
  • dtype: numpy data type of the tensor elements
  • tensor: Name of the associated tensor for use with graph.get_tensor_by_name()

tfjs_graph_converter.util.get_input_nodes

Return a list of NodeInfo tuples describing a graph's inputs.

Arguments

  • graph: TF Graph or GraphDef object

Returns

List of NodeInfo tuples with basic properties of the graph's inputs.

Example

import tensorflow as tf
import tfjs_graph_converter as tfjs

# load a frozen model graph, assume the input is a single image tensor
with tf.io.gfile.GFile("frozen_model.pb", "rb") as f:
    graph_def = tf.compat.v1.GraphDef()
    graph_def.ParseFromString(f.read())

# get information about the input
model_input = tfjs.util.get_input_nodes(graph_def)[0]
input_shape = model_input.shape
# extract the width and height of the input dimension (assume "NHWC"-format)
target_size = input_shape[len(input_shape)-3,-1] if len(input_shape) in (3, 4) else None

# Load an input image according to the dimensions and element type of the model input
img = tf.keras.preprocessing.image.load_img("image.png", target_size=target_size)
x = tf.keras.preprocessing.image.img_to_array(img, dtype=model_input.dtype)
# scale pixels to [0..1]
x /= 255.0
# extend dimensions if required
if len(input_shape) > 3:
    x = x[tf.newaxis, ...]

# "x" is now a suitable input for the model

tfjs_graph_converter.util.get_input_tensors

Return a list of input tensor names for use with graph.get_tensor_by_name().

Arguments

  • graph: TF Graph or GraphDef object

Returns

List of strings containing input tensor names.

Examples

import tensorflow as tf
import tfjs_graph_converter as tfjs

# load a frozen model graph, assume the input is a single image tensor
with tf.io.gfile.GFile("frozen_model.pb", "rb") as f:
    graph_def = tf.compat.v1.GraphDef()
    graph_def.ParseFromString(f.read())

# load suitable model input - see get_input_nodes() for an example
image_data = ...

# create a graph
with tf.Graph().as_default() as graph:
    tf.import_graph_def(graph_def, name="")

# create a session
with tf.compat.v1.Session(graph=graph) as sess:
    output_tensors = tfjs.util.get_output_tensors(graph)
    input_name = tfjs.util.get_input_tensors(graph)[0]
    input_tensor = graph.get_tensor_by_name(input_name)
    # run the model
    results = sess.run(output_tensors, feed_dict={input_tensor: image_data})
# ...

tfjs_graph_converter.util.get_output_nodes

Return a list of NodeInfo tuples describing a graph's outputs.

Arguments

  • graph: TF Graph or GraphDef object

Returns

List of NodeInfo tuples with basic properties of the graph's outputs. Note that the shape field is not being populated and remains an empty list.

tfjs_graph_converter.util.get_output_tensors

Return a list of input tensor names for use with graph.get_tensor_by_name(). This is useful for passing output tensor names to Session.run().

Arguments

  • graph: TF Graph or GraphDef object

Returns

List of strings containing output tensor names.

Example

  • see get_input_tensors() for a usage example

tfjs_graph_converter.version

Contains the module version.

tfjs_graph_converter.version.VERSION

String that contains the module version.

tfjs_graph_converter.common

Constants used throughout the module.