-
80-NL315-14 A
MAY CONTAIN U.S. AND INTERNATIONAL EXPORT CONTROLLED INFORMATION
|
Snapdragon Neural Processing Engine SDK
Reference Guide
|
This chapter covers the conversion tools used to convert a TensorFlow model to a SNPE DLC file. The DLC file is used by the SNPE runtime for execution of the neural network.
Prerequisites
A trained TensorFlow model saved as either:
SNPE like many other neural network runtime engines uses layers as building blocks to define the structure of neural networks. TensorFlow on the other hand, defines a neural network as a graph of nodes and a layer is defined as a set of nodes within the graph.
With this in mind, in order to properly convert a TensorFlow graph into a SNPE DLC file the following requirements must be met when defining a TensorFlow graph:
More information about graph compatibility can be found in the TensorFlow Graph Compatibility chapter.
The snpe-tensorflow-to-dlc tool converts a TensorFlow model into an equivalent SNPE DLC file. The options available in the tool are shown below:
$ snpe-tensorflow-to-dlc -h
usage: snpe-tensorflow-to-dlc [-h] [--input_network INPUT_NETWORK]
[-o OUTPUT_PATH]
[--copyright_file COPYRIGHT_FILE]
[--model_version MODEL_VERSION]
[--disable_batchnorm_folding]
[--input_type INPUT_NAME INPUT_TYPE]
[--input_encoding INPUT_NAME INPUT_ENCODING]
[--validation_target RUNTIME_TARGET PROCESSOR_TARGET]
[--strict] [--debug [DEBUG]] -d INPUT_NAME
INPUT_DIM --out_node OUT_NODE
[--allow_unconsumed_nodes]
Script to convert tensorflowmodel into a DLC file.
optional arguments:
-h, --help show this help message and exit
required arguments:
--input_network INPUT_NETWORK, -i INPUT_NETWORK
Path to the source framework model.
-d INPUT_NAME INPUT_DIM, --input_dim INPUT_NAME INPUT_DIM
The names and dimensions of the network input layers
specified in the format "input_name" comma-separated-
dimensions, for example: "data" 1,224,224,3. Note that
the quotes should always be included in order to
handle special characters, spaces, etc. For multiple
inputs specify multiple --input_dim on the command
line like: --input_dim "data1" 1,224,224,3 --input_dim
"data2" 1,50,100,3.
--out_node OUT_NODE Name of the graph's output node.
optional arguments:
-o OUTPUT_PATH, --output_path OUTPUT_PATH
Path where the converted Output model should be
saved.If not specified, the converter model will be
written to a file with same name as the input model
--copyright_file COPYRIGHT_FILE
Path to copyright file. If provided, the content of
the file will be added to the output model.
--model_version MODEL_VERSION
User-defined ASCII string to identify the model, only
first 64 bytes will be stored
--disable_batchnorm_folding
If not specified, converter will try to fold batchnorm
into previous convolution layer
--input_type INPUT_NAME INPUT_TYPE, -t INPUT_NAME INPUT_TYPE
Type of data expected by each input op/layer. Type for
each input is |default| if not specified. For example:
"data" image.Note that the quotes should always be
included in order to handle special characters,
spaces,etc. For multiple inputs specify multiple
--input_type on the command line. Eg: --input_type
"data1" image --input_type "data2" opaque These
options get used by DSP runtime and following
descriptions state how input will be handled for each
option. Image: input is float between 0-255 and the
input's mean is 0.0f and the input's max is 255.0f. We
will cast the float to uint8ts and pass the uint8ts to
the DSP. Default: pass the input as floats to the dsp
directly and the DSP will quantize it. Opaque: assumes
input is float because the consumer layer(i.e next
layer) requires it as float, therefore it won't be
quantized.Choices supported:['image', 'default',
'opaque']
--input_encoding INPUT_NAME INPUT_ENCODING, -e INPUT_NAME INPUT_ENCODING
Image encoding of the source images. Default is bgr.
Eg usage: "data" rgba Note the quotes should always be
included in order to handle special characters,
spaces, etc. For multiple inputs specify
--input_encoding for each on the command line. Eg:
--input_encoding "data1" rgba --input_encoding "data2"
other. Use options: color encodings(bgr,rgb, nv21...)
if input is image; time_series: for inputs of rnn
models; other: if input doesn't follow above
categories or is unknown. Choices supported:['bgr',
'rgb', 'rgba', 'argb32', 'nv21', 'time_series',
'other']
--validation_target RUNTIME_TARGET PROCESSOR_TARGET
A combination of processor and runtime target against
which model will be validated. Choices for
RUNTIME_TARGET: {cpu, gpu, dsp}. Choices for
PROCESSOR_TARGET: {snapdragon_801, snapdragon_820,
snapdragon_835}.If not specified, will validate model
against {snapdragon_820, snapdragon_835} across all
runtime targets.
--strict If specified, will validate in strict mode whereby
model will not be produced if it violates constraints
of the specified validation target. If not specified,
will validate model in permissive mode against the
specified validation target.
--debug [DEBUG] Run the converter in debug mode.
--allow_unconsumed_nodes
Uses a relaxed graph node to layer mapping algorithm
which may not use all graph nodes during conversion
while retaining structural integrity.
This section details the usage of the converter arguments.
input_network argument:
input_dim argument:
out_node argument:
output_path argument:
$ snpe-tensorflow-to-dlc --input_network inception_v3.pb --input_dim 1,299,299,3 --output_path net.dlc --in_node input --out_node InceptionV3/Predictions/Reshape_1 --allow_unconsumed_nodes --verbose 2017-04-01 20:23:42,794 - 316 - INFO - ============================================================== 2017-04-01 20:23:42,794 - 317 - INFO - Building Network 2017-04-01 20:23:42,794 - 318 - INFO - ============================================================== 2017-04-01 20:23:42,794 - 338 - INFO - Building layer (INPUT) with nodes: Mul, shape [1, 299, 299, 3] 2017-04-01 20:23:42,794 - 348 - INFO - Converting scope (conv): [u'conv/Conv2D', u'conv/batchnorm', u'conv'] 2017-04-01 20:23:43,542 - 348 - INFO - Converting scope (conv_1): [u'conv_1/Conv2D', u'conv_1/batchnorm', u'conv_1'] ... Building layer (Convolution) with nodes: [u'convolution1/Conv2D', u'convolution1/BiasAdd'] Building layer (Convolution) with nodes: [u'convolution2/Conv2D', u'convolution2/BiasAdd'] Building layer (ElementWiseSum) with nodes: [u'output/Add'] Model conversion completed!
The converted model can be viewed using the snpe-dlc-info tool. A subset of a sample output is shown below.
$ snpe-dlc-info -i <path_to_dlc>/net.dlc DLC info for <path_to_dlc>/net.dlc Model Version: ----------------------------------------------------------------------------------------------------------------------- | Id | Name | Type | Inputs | Outputs | Out Dims | Parameters | ----------------------------------------------------------------------------------------------------------------------- | 0 | Mul:0 | data | Mul:0 | Mul:0 | 299x299x3 | input_preprocessing: passthrough | | 1 | conv/Conv2D | convolutional | Mul:0 | conv/batchnorm:0 | 149x149x32 | padding x: 0 | | | | | | | | padding y: 0 | | | | | | | | stride x: 2 | | | | | | | | stride y: 2 | | | | | | | | num filters: 32 | | | | | | | | kernel: 3x3 | | | | | | | | param count: 896 (0.00376%) | | | | | | | | MACs: 19M (0.336%) | ...
80-NL315-14 A
MAY CONTAIN U.S. AND INTERNATIONAL EXPORT CONTROLLED INFORMATION