Graph.allocate_with_fifos()

Info	Value
Package	mvnc
Module	mvncapi
Version	2.0
See also	Device, Graph, Fifo, Graph.allocate()

Overview

This method allocates a network graph to the device and creates and allocates two associated Fifos for graph input and output. This is a convenient way to set up a Graph for inference without setting up Fifos explicitly. Each Graph instance can only be allocated to a single Device instance.

Upon successful return from this method, the Graph will be ready to perform inferences and the input and output Fifos will be ready for reading/writing.

Calling Graph.destroy() for the Graph and Fifo.destroy() for each Fifo frees the memory allocated with this method.

Syntax

input_fifo, output_fifo = graph.allocate_with_fifos(
        device, graph_buffer,
        input_fifo_type=mvncapi.FifoType.HOST_WO, 
        input_fifo_num_elem=2,
        input_fifo_data_type=mvncapi.FifoDataType.FP32,
        output_fifo_type=mvncapi.FifoType.HOST_RO, 
        output_fifo_num_elem=2,
        output_fifo_data_type=mvncapi.FifoDataType.FP32
)

Parameters

Parameter	Type	Default	Description
device	Device		A Device that this Graph will be allocated to. The DeviceState must be OPENED.
graph_buffer	bytes		Raw data read from a neural network graph file that was created with the Intel® Movidius™ Neural Compute SDK mvNCCompile tool.
input_fifo_type	int	FifoType.HOST_WO	A member of the FifoType enumeration, which sets read/write access to the Fifo.
input_fifo_num_elem	int	2	The maximum number of elements that the input Fifo can hold.
input_fifo_data_type	int	FifoDataType.FP32	A member of the FifoDataType enumeration.
output_fifo_type	int	FifoType.HOST_RO	A member of the FifoType enumeration, which sets read/write access to the Fifo.
output_fifo_num_elem	int	2	The maximum number of elements that the output Fifo can hold.
output_fifo_data_type	int	FifoDataType.FP32	A member of the FifoDataType enumeration.

Return

input_fifo, output_fifo

The input and output Fifos that were created and allocated by this method.

Raises

Exception with a status code from Status if underlying function calls return a status other than Status.OK.

Notes

This method takes the place of explicit calls for Graph.allocate(), Fifo creation, and Fifo.allocate().

Example

from mvnc import mvncapi

# Get a list of valid device identifiers
device_list = mvncapi.enumerate_devices()

# Create a Device instance for the first device found
device = mvncapi.Device(device_list[0])

# Open communication with the device
device.open()

# Create a Graph
graph = mvncapi.Graph('graph1')

# Read a compiled network graph from file (set the graph_filepath correctly for your graph file)
graph_filepath = './graph'
with open(graph_filepath, 'rb') as f:
    graph_buffer = f.read()

# Allocate the graph on the device
input_fifo, output_fifo = graph.allocate_with_fifos(device, graph_buffer)

#
# Use the device...
#

# Deallocate and destroy the graph handle and fifo handles, close the device, and destroy the device handle
input_fifo.destroy()
output_fifo.destroy()
graph.destroy()
device.close()
device.destroy()