Intel-Movidius-NCS-Keras

Arda Mavi

Runing Keras(Background: TensorFlow) with Intel Movidius Neural Compute Stick

The under construction!

Intel Movidius Neural Compute Stick

Thank Mustafa Aldemir and Intel Corporation Turkey very much for this gift !

About Intel Movidius Neural Compute Stick:

Official Web Page: developer.movidius.com

Installing SDK command:

git clone http://github.com/Movidius/ncsdk && cd ncsdk && make install

NCS Examples:

Installing NCS SDK's Examples command:

Run make examples command inside ncsdk folder.

For more NCS examples:

Look up: github.com/movidius/ncappzoo

Keras Model to NCS Graph:

File: keras2graph.py

Converts Keras model and weights to Intel Movidius technology internal compiled format.

keras2graph process turn used TensorFlow backend Keras models to NCS graph.

But how ?

Keras Model -> TensorFlow Session -> TensorFlow Meta File -> NCS Graph

Run Command:

python3 keras2graph.py <Keras_Model> <Model_input_layer_name> <Model_output_layer_name> <Keras_Model_Weights> <Output_Graph>

This command creates and saves graph file as <Output_Graph>.

Example Run Command:

python3 keras2graph.py Keras_Model/model.json input_1 activation_7/Softmax Keras_Model/weights.h5 ./graph

Functions:

• keras_to_graph
• get_keras_model
• keras_to_tf
• tf_to_graph

keras_to_graph

keras_to_graph(model_path, model_in, model_out, weights_path, graph_path, take_tf_files = False)

For converting Keras Model to Movidius NCS graph.

Arguments:

Inputs:

• model_path: Model location. Json file. Data Type: String
• model_in: Name of model's input layer. Data Type: String
• model_out: Name of model's output layer. Data Type: String
• graph_path: Location of output Intel Movidius technology internal compiled format graph. Data Type: String
• take_tf_files: If you don't want to delete created TensorFlow model files(meta file inside), select False. Data Type: Bool

get_keras_model

get_keras_model(model_path, weights_path)

For getting Keras model.

Arguments:

Inputs:

• model_path: Keras model Json file. Data Type: String
• weights_path: Weights of Keras model. h5 file. Data Type: String

Outputs:

• model: Keras model object.

keras_to_tf

keras_to_tf(tf_model_path)

For converting Keras model to TensorFlow session and saving TensorFlow graph to meta file.

Important Note: Use this function after get_keras_model function.

Arguments:

Inputs:

• tf_model_path: Saving location of output meta file. Data Type: String

tf_to_graph

tf_to_graph(tf_model_path, model_in, model_out, graph_path):

For converting TensorFlow graph to Intel Movidius technology internal compiled format.

Arguments:

Inputs:

• tf_model_path: Location of TensorFlow meta file. Data Type: String
• model_in: Name of model's input layer. Data Type: String
• model_out: Name of model's output layer. Data Type: String
• graph_path: Location of output Intel Movidius technology internal compiled format graph. Data Type: String

NCS Process:

File: ncs_process.py

Functions:

All in one functions

• ready_ai_ncs
• release_ai_ncs

Functions

• ncs_predict

All in one functions:

ready_ai_ncs

ready_ai_ncs(graph_path, device_index=0)

For getting ready Movidius NCS with model to use with one command.

Arguments:

Inputs:

• graph_path: Graph file path. Data type: String
• device_index: Which device you want use. Data type: Int , Default value: 0

Outputs:

• ncs_model: Model object. This object ready for running graph in NCS (ncs_model object used with ncs_predict() function)
• device: Using device object.

release_ai_ncs

release_ai_ncs(ncs_model, device)

For release model(in the Movidius NCS) and close Movidius device.

Arguments:

Inputs:

• ncs_model: Model object. (Returned from ready_ai_ncs)
• device: Using device object.

Functions:

ncs_predict

ncs_predict(ncs_model, inputs)

For running Intel Movidius technology internal compiled format graph in Intel Movidius NCS.

Arguments:

Inputs:

• ncs_model: Intel Movidius technology internal compiled format graph object.
• inputs: Input matrics. Data type: Numpy Array

Outputs:

• outputs: Output of graph. Data type: Numpy Array

Important Notes:

• Install necessary modules with sudo pip3 install -r requirements.txt command.