背景资料：
我写了一个TensorFlow模型，非常类似于TensorFlow提供的premade iris classification model。差异相对较小：
我现在想要将模型导出为.tflite文件。但根据TensorFlow Developer Guide，我需要首先将模型导出到tf.GraphDef文件，然后冻结它，然后我才能转换它。但是，TensorFlow提供的用于从自定义模型创建tutorial文件的.pb似乎仅针对图像分类模型进行了优化。
题：
那么如何将像虹膜分类示例模型这样的模型转换为.tflite文件呢？是否有更简单，更直接的方法，而不必将其导出到.pb文件，然后冻结等等？基于虹膜分类代码的示例或指向更明确的教程的链接将非常有用！
其他信息：
码：
可以通过输入以下命令克隆虹膜分类代码：
git clone https://github.com/tensorflow/models
但是如果你不想下载整个软件包，这里是：
这是名为premade_estimator.py的分类器文件：
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""An Example of a DNNClassifier for the Iris dataset."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import tensorflow as tf
import iris_data
parser = argparse.ArgumentParser()
parser.add_argument('--batch_size', default=100, type=int, help='batch size')
parser.add_argument('--train_steps', default=1000, type=int,
help='number of training steps')
def main(argv):
args = parser.parse_args(argv[1:])
# Fetch the data
(train_x, train_y), (test_x, test_y) = iris_data.load_data()
# Feature columns describe how to use the input.
my_feature_columns = []
for key in train_x.keys():
my_feature_columns.append(tf.feature_column.numeric_column(key=key))
# Build 2 hidden layer DNN with 10, 10 units respectively.
classifier = tf.estimator.DNNClassifier(
feature_columns=my_feature_columns,
# Two hidden layers of 10 nodes each.
hidden_units=[10, 10],
# The model must choose between 3 classes.
n_classes=3)
# Train the Model.
classifier.train(
input_fn=lambda: iris_data.train_input_fn(train_x, train_y,
args.batch_size),
steps=args.train_steps)
# Evaluate the model.
eval_result = classifier.evaluate(
input_fn=lambda: iris_data.eval_input_fn(test_x, test_y,
args.batch_size))
print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))
# Generate predictions from the model
expected = ['Setosa', 'Versicolor', 'Virginica']
predict_x = {
'SepalLength': [5.1, 5.9, 6.9],
'SepalWidth': [3.3, 3.0, 3.1],
'PetalLength': [1.7, 4.2, 5.4],
'PetalWidth': [0.5, 1.5, 2.1],
}
predictions = classifier.predict(
input_fn=lambda: iris_data.eval_input_fn(predict_x,
labels=None,
batch_size=args.batch_size))
template = '\nPrediction is "{}" ({:.1f}%), expected "{}"'
for pred_dict, expec in zip(predictions, expected):
class_id = pred_dict['class_ids'][0]
probability = pred_dict['probabilities'][class_id]
print(template.format(iris_data.SPECIES[class_id],
100 * probability, expec))
if __name__ == '__main__':
# tf.logging.set_verbosity(tf.logging.INFO)
tf.app.run(main)
这是名为iris_data.py的数据文件：
import pandas as pd
import tensorflow as tf
TRAIN_URL = "http://download.tensorflow.org/data/iris_training.csv"
TEST_URL = "http://download.tensorflow.org/data/iris_test.csv"
CSV_COLUMN_NAMES = ['SepalLength', 'SepalWidth',
'PetalLength', 'PetalWidth', 'Species']
SPECIES = ['Setosa', 'Versicolor', 'Virginica']
def maybe_download():
train_path = tf.keras.utils.get_file(TRAIN_URL.split('/')[-1], TRAIN_URL)
test_path = tf.keras.utils.get_file(TEST_URL.split('/')[-1], TEST_URL)
return train_path, test_path
def load_data(y_name='Species'):
"""Returns the iris dataset as (train_x, train_y), (test_x, test_y)."""
train_path, test_path = maybe_download()
train = pd.read_csv(train_path, names=CSV_COLUMN_NAMES, header=0)
train_x, train_y = train, train.pop(y_name)
test = pd.read_csv(test_path, names=CSV_COLUMN_NAMES, header=0)
test_x, test_y = test, test.pop(y_name)
return (train_x, train_y), (test_x, test_y)
def train_input_fn(features, labels, batch_size):
"""An input function for training"""
# Convert the inputs to a Dataset.
dataset = tf.data.Dataset.from_tensor_slices((dict(features), labels))
# Shuffle, repeat, and batch the examples.
dataset = dataset.shuffle(1000).repeat().batch(batch_size)
# Return the dataset.
return dataset
def eval_input_fn(features, labels, batch_size):
"""An input function for evaluation or prediction"""
features = dict(features)
if labels is None:
# No labels, use only features.
inputs = features
else:
inputs = (features, labels)
# Convert the inputs to a Dataset.
dataset = tf.data.Dataset.from_tensor_slices(inputs)
# Batch the examples
assert batch_size is not None, "batch_size must not be None"
dataset = dataset.batch(batch_size)
# Return the dataset.
return dataset
**更新**
好的，我发现了一段看似非常有用的代码on this page：
import tensorflow as tf
img = tf.placeholder(name="img", dtype=tf.float32, shape=(1, 64, 64, 3))
val = img + tf.constant([1., 2., 3.]) + tf.constant([1., 4., 4.])
out = tf.identity(val, name="out")
with tf.Session() as sess:
tflite_model = tf.contrib.lite.toco_convert(sess.graph_def, [img], [out])
open("test.tflite", "wb").write(tflite_model)
这个小家伙直接将简单模型转换为TensorFlow Lite模型。现在，我所要做的就是找到一种方法使其适应虹膜分类模型。有什么建议？
有没有更简单，更直接的方法，而不必将其导出到.pb文件，然后冻结它等等？
是的，正如您在更新的问题中指出的那样，有可能freeze the graph并直接在python api中使用toco_convert。它需要冻结图形并确定输入和输出形状。在您的问题中，没有冻结图步骤，因为没有变量。如果你有变量并运行toco而不先将它们转换为常量，那么toco会抱怨！
现在，我所要做的就是找到一种方法使其适应虹膜分类模型。有什么建议？
这个有点棘手，需要更多的工作。基本上，您需要加载图形并找出输入和输出张量名称，然后冻结图形并调用toco_convert。为了在这种情况下找到输入和输出张量名称（你没有定义图形），你必须围绕生成的图形，并根据输入的形状，名称等确定它们。这是你可以附加的代码premade_estimator.py中主函数的结束，在这种情况下生成tflite图。
print("\n====== classifier model_dir, latest_checkpoint ===========")
print(classifier.model_dir)
print(classifier.latest_checkpoint())
debug = False
with tf.Session() as sess:
# First let's load meta graph and restore weights
latest_checkpoint_path = classifier.latest_checkpoint()
saver = tf.train.import_meta_graph(latest_checkpoint_path + '.meta')
saver.restore(sess, latest_checkpoint_path)
# Get the input and output tensors needed for toco.
# These were determined based on the debugging info printed / saved below.
input_tensor = sess.graph.get_tensor_by_name("dnn/input_from_feature_columns/input_layer/concat:0")
input_tensor.set_shape([1, 4])
out_tensor = sess.graph.get_tensor_by_name("dnn/logits/BiasAdd:0")
out_tensor.set_shape([1, 3])
# Pass the output node name we are interested in.
# Based on the debugging info printed / saved below, pulled out the
# name of the node for the logits (before the softmax is applied).
frozen_graph_def = tf.graph_util.convert_variables_to_constants(
sess, sess.graph_def, output_node_names=["dnn/logits/BiasAdd"])
if debug is True:
print("\nORIGINAL GRAPH DEF Ops ===========================================")
ops = sess.graph.get_operations()
for op in ops:
if "BiasAdd" in op.name or "input_layer" in op.name:
print([op.name, op.values()])
# save original graphdef to text file
with open("estimator_graph.pbtxt", "w") as fp:
fp.write(str(sess.graph_def))
print("\nFROZEN GRAPH DEF Nodes ===========================================")
for node in frozen_graph_def.node:
print(node.name)
# save frozen graph def to text file
with open("estimator_frozen_graph.pbtxt", "w") as fp:
fp.write(str(frozen_graph_def))
tflite_model = tf.contrib.lite.toco_convert(frozen_graph_def, [input_tensor], [out_tensor])
open("estimator_model.tflite", "wb").write(tflite_model)
注意：我假设从最后一层（应用Softmax之前）的logits作为输出，对应于节点dnn / logits / BiasAdd。如果你想要概率，我相信它是dnn / head / predictions / probability。
这是一种更标准的方法，而不是使用toco_convert。感谢Pannag Sanketi提供了基于toco的示例，该示例是此代码的基础。
请注意，输出层是logits，因为我们使用的是分类NN。如果我们有一个回归NN，它将是不同的。 classifier是你建立的NN模型。
def export_tflite(classifier):
with tf.Session() as sess:
# First let's load meta graph and restore weights
latest_checkpoint_path = classifier.latest_checkpoint()
saver = tf.train.import_meta_graph(latest_checkpoint_path + '.meta')
saver.restore(sess, latest_checkpoint_path)
# Get the input and output tensors
input_tensor = sess.graph.get_tensor_by_name("dnn/input_from_feature_columns/input_layer/concat:0")
out_tensor = sess.graph.get_tensor_by_name("dnn/logits/BiasAdd:0")
# here the code differs from the toco example above
sess.run(tf.global_variables_initializer())
converter = tf.lite.TFLiteConverter.from_session(sess, [input_tensor], [out_tensor])
tflite_model = converter.convert()
open("converted_model.tflite", "wb").write(tflite_model)