This is an keras and tensorflow example to show the difference of a model's execution speed between keras and tensorflow, executed from python and c++.
This code forked from JackyTung/tensorgraph.
More description can check on JackyTung's blog post
- tensorflow installation, https://www.tensorflow.org/
go to "GET STARTED" --> "installing from source" - bazel installation, http://www.bazel.io/docs/install.html
How to generate checkpoint, graph.pb, tensorboard.
The directory struct is
mnist_tf.py
mnist_keras_tf.py
mnist_keras.py
mnist_tf_conv.py
mnist_keras_tf_conv.py
mnist_keras_conv.py
board/
After run
$ python mnist_tf.py
The directory struct will be expected to
mnist_tf.py
...
board/
event.out.tfevents
models_tf/
graph.pb
model.ckpt
Mnist_data/
...
From Tensorflow official guide says that:
What this does is load the GraphDef, pull in the values for all the variables from the latest checkpoint file, and then replace each Variable op with a Const that has the numerical data for the weights stored in its attributes It then strips away all the extraneous nodes that aren't used for forward inference, and saves out the resulting GraphDef into an output file
Hence, we do the following steps to generate frozen graph
bazel build tensorflow/python/tools:freeze_graph && \
bazel-bin/tensorflow/python/tools/freeze_graph \
--input_graph=graph.pb \
--input_checkpoint=model.ckpt \
--output_graph=/tmp/frozen_graph.pb --output_node_names=softmax
How to load graph with tensorflow c++ api and do the prediction.
Put the directory to tensorflow source code.
Here is the final directory structure:
tensorflow/tensorflow/loadgraph
tensorflow/tensorflow/loadgraph/mnist.cc
tensorflow/tensorflow/loadgraph/MNIST.h
tensorflow/tensorflow/loadgraph/BUILD
Compile and Run
From inside the project folder call $bazel build :mnistpredict
From the repository root, go into bazel-bin/tensorflow/loadgraph.
Copy the frozen_graph.pb and Mnist_data to bazel-bin/tensorflow/loadgraph
Then run ./mnistpredict and check the output