jevois::dnn::Network Class Reference

#include <jevois/DNN/Network.H>

Abstract class to represent a neural network.

Derived classes provide implementation via OpenCV (on CPU, OpenCL, or OpenVino/Myriad-X), Amlogic/Vivante NPU, Hailo-8, Python, or Google Coral TPU.

The Network class can apply optional transforms to the output tensors:

• dequant: dequantize the output tensors
• outtransform: apply a semicolon-separated sequence (from left to right) of transforms to the outputs. Outputs are numbered starting at 0. Available transforms are:
  • shape(outnum, AxBxC...) to reshape an output to a given new shape. Does not change or move any tensor data. Total number of data elements must remain the same. Useful to squeeze/unsqueeze tensors.
  • transpose(outnum, oldaxisA, oldaxisB, ...) where transposed new axis 0 (the outermost dimension, typically batch size) will be from oldaxisA, new axis 1 from oldaxisB, etc. If outnum is *, transpose all outputs.
  • order(oldidx0, oldidx1, ...) where the new ordering of the output tensors will be: new tensor 0: old tensor oldidx0 (which is zero-based); new tensor 1: old tensor oldidx1, etc. It is ok to have duplicated or missing entries.
  • split(outnum, axis, newsize1, ..., newsizeN) where axis 0 is the outtermost dimension (typically, batch size), and newsize1 + ... + newsizeN must be equal to the original size of that axis. If outnum is *, split all outputs.
  • merge(axis, outnum1, ..., outnumN) where axis 0 is the outermost dimension (typically, batch size) and outnum1, ..., outnumN are the outputs to merge along that axis. All the outputs to be merged must have matching number of dimensions, and matching sizes on all other axes. The merged tensor will replace the first output listed in the merge, and the other listed will be removed. Outputs to merge must be listed in ascending order (use an order() transform first if needed)

See the model zoo files in /jevoispro/share/dnn/ for examples. For instance:

outtransform: "split(*,1,80,64); order(1,0,3,2,5,4); transpose(*,0,2,3,1)"

Definition at line 210 of file Network.H.

Inheritance diagram for jevois::dnn::Network:

Collaboration diagram for jevois::dnn::Network:

Public Member Functions
virtual	~Network ()
	Destructor.
void	waitBeforeDestroy ()
	If network is currently loading, wait until that is done before destroying.
bool	ready ()
	Returns true when network is ready to run (loaded and initialized)
virtual std::vector< vsi_nn_tensor_attr_t >	inputShapes ()=0
	Get shapes of all input tensors.
virtual std::vector< vsi_nn_tensor_attr_t >	outputShapes ()=0
	Get shapes of all output tensors.
std::vector< cv::Mat >	process (std::vector< cv::Mat > const &blobs, std::vector< std::string > &info)
	Process input blobs and obtain output blobs.
virtual void	freeze (bool doit)
	Freeze/unfreeze parameters that users should not change while running.
Public Member Functions inherited from jevois::Component
	Component (std::string const &instance)
	Constructor.
virtual	~Component ()
	Virtual destructor for safe inheritance.
template<class Comp , typename... Args>
std::shared_ptr< Comp >	addSubComponent (std::string const &instance, Args &&...args)
	Pseudo-constructor: construct and add another component as a subcomponent of this one.
template<class Comp >
void	removeSubComponent (std::shared_ptr< Comp > &component)
	Remove a sub-Component from this Component, by shared_ptr.
void	removeSubComponent (std::string const &instance, bool warnIfNotFound=true)
	Remove a sub-Component from this Component, by instance name.
template<class Comp = jevois::Component>
std::shared_ptr< Comp >	getSubComponent (std::string const &instance) const
	Get a sub-component by instance name.
bool	isTopLevel () const
	Returns true if this component is top-level, i.e., its parent is jevois::Manager.
bool	initialized () const
	Has this component been initialized yet?
std::string const &	className () const
	The class name of this component.
std::string const &	instanceName () const
	The instance name of this component.
template<typename T >
std::vector< std::string >	setParamVal (std::string const &paramdescriptor, T const &val)
	Set a parameter value.
template<typename T >
void	setParamValUnique (std::string const &paramdescriptor, T const &val)
	Set a parameter value, simple version assuming only one parameter match.
template<typename T >
std::vector< std::pair< std::string, T > >	getParamVal (std::string const &paramdescriptor) const
	Get parameter(s) value(s) by descriptor.
template<typename T >
T	getParamValUnique (std::string const &paramdescriptor) const
	Get a parameter value, simple version assuming only one parameter match.
std::vector< std::string >	setParamString (std::string const &paramdescriptor, std::string const &val)
	Set a parameter value, by string.
void	setParamStringUnique (std::string const &paramdescriptor, std::string const &val)
	Set a parameter value by string, simple version assuming only one parameter match.
std::vector< std::pair< std::string, std::string > >	getParamString (std::string const &paramdescriptor) const
	Get a parameter value, by string.
std::string	getParamStringUnique (std::string const &paramdescriptor) const
	Get a parameter value by string, simple version assuming only one parameter match.
void	freezeParam (std::string const &paramdescriptor, bool doit)
	Freeze/unfreeze a parameter, by name, see ParameterBase::freeze()
void	freezeAllParams (bool doit)
	Freeze all parameters.
std::string	descriptor () const
	Get our full descriptor (including all parents) as [Instancename]:[...]:[...].
void	setParamsFromFile (std::string const &filename)
	Set some parameters from a file.
std::istream &	setParamsFromStream (std::istream &is, std::string const &absfile)
	Set some parameters from an open stream.
virtual void	paramInfo (std::shared_ptr< UserInterface > s, std::map< std::string, std::string > &categs, bool skipFrozen, std::string const &cname="", std::string const &pfx="")
	Get machine-oriented descriptions of all parameters.
void	foreachParam (std::function< void(std::string const &compname, ParameterBase *p)> func, std::string const &cname="")
	Run a function on every param we hold.
template<typename T >
std::shared_ptr< DynamicParameter< T > >	addDynamicParameter (std::string const &name, std::string const &description, T const &defaultValue, ParameterCategory const &category)
	Add a new parameter after the Component has already been constructed.
template<typename T , template< typename > class ValidValuesSpec>
std::shared_ptr< DynamicParameter< T > >	addDynamicParameter (std::string const &name, std::string const &description, T const &defaultValue, ValidValuesSpec< T > const &validValuesSpec, ParameterCategory const &category)
	Add a new parameter after the Component has already been constructed.
template<typename T >
void	setDynamicParameterCallback (std::string const &name, std::function< void(T const &)> cb, bool callnow=true)
	Register a callback with a previously created dynamic parameter.
void	removeDynamicParameter (std::string const &name, bool throw_if_not_found=true)
	Remove a previously added dynamic parameter.
void	setPath (std::string const &path)
	Assign a filesystem path to this component.
std::filesystem::path	absolutePath (std::filesystem::path const &path="")
	If given path is relative (not starting with /), prepend the Component path to it.
Public Member Functions inherited from jevois::ParameterRegistry
virtual	~ParameterRegistry ()
	Virtual destructor for safe inheritance.

Protected Member Functions
virtual void	load ()=0
	Load from disk.
virtual std::vector< cv::Mat >	doprocess (std::vector< cv::Mat > const &blobs, std::vector< std::string > &info)=0
	Process input blobs and obtain output blobs.
void	onParamChange (network::outtransform const &param, std::string const &val) override
Protected Member Functions inherited from jevois::Component
virtual void	preInit ()
	Called before all sub-Components are init()ed.
virtual void	postInit ()
	Called after all sub-Components are init()ed.
virtual void	preUninit ()
	Called before all sub-Components are uninit()ed.
virtual void	postUninit ()
	Called after all sub-Components are uninit()ed.
Protected Member Functions inherited from jevois::ParameterRegistry
void	addParameter (ParameterBase *const param)
	The Parameter class uses this method to register itself on construction with its owning Component.
void	removeParameter (ParameterBase *const param)
	The Parameter class uses this method to un-register itself on destruction with its owning Component.
void	callbackInitCall ()
	For all parameters that have a callback which has never been called, call it with the default param value.

Related Symbols
(Note that these are not member symbols.)
	JEVOIS_DECLARE_PARAMETER (comment, std::string, "Optional comment about the network", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (url, std::string, "Optional URL for the network", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (dataroot, std::string, "Root directory to use when config or model parameters " "are relative paths.", JEVOIS_SHARE_PATH, ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (config, std::string, "Path to a text file that contains network configuration. " "Can have extension .prototxt (Caffe), .pbtxt (TensorFlow), or .cfg (Darknet). " "If path is relative, it will be prefixed by dataroot.", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (model, std::string, "Path to a binary file of model contains trained weights. " "Can have extension .caffemodel (Caffe), .pb (TensorFlow), .t7 or .net (Torch), " ".tflite (TensorFlow Lite), or .weights (Darknet). If path is relative, it will be " "prefixed by dataroot.", "", ParamCateg)
	Parameter.
	JEVOIS_DEFINE_ENUM_CLASS (Target,(CPU)(OpenCL)(OpenCL_FP16)(Myriad)(NPU))
	Enum.
	JEVOIS_DECLARE_PARAMETER (target, Target, "OpenCV compute target to use. Changes will take effect " "next time you load a different model.", Target::CPU, Target_Values, ParamCateg)
	Parameter.
	JEVOIS_DEFINE_ENUM_CLASS (Backend,(OpenCV)(InferenceEngine)(TimVX))
	Enum.
	JEVOIS_DECLARE_PARAMETER (backend, Backend, "OpenCV compute backend to use. Default will use the inference " "engine if available, otherwise OpenCV (note that inference engine only works on Intel " "processors or MyriadX hardware, thus you should normally select OpenCV when running " "on JeVois-Pro Platform, unless you want to use an optional MyriadX accelerator). " "Changes will take effect next time you load a model.", JEVOIS_BACKEND_DEFAULT, Backend_Values, ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (intensors, std::string, "Specification of input tensors", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (extraintensors, std::string, "Specification of extra fixed input tensors that will be " "added after the regular intensors. Format is: " "<type>:<shape>:val1 val2 ... valN, <type>:<shape>:val1 ... valN. For example, for " "URetinex-Net: 32F:1x1x1:3.0", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (outtensors, std::string, "Specification of output tensors", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER_WITH_CALLBACK (outtransform, std::string, "Specification of a sequence of operations " "(reshape, transpose, split, merge, etc) to be applied to the output " "tensors. Useful to match a particular network's outputs to the " "tensor shapes expected by a particular post-processor. The sequence of " "operations is separated by semicolons, and applied from left to right. " "Operations specified here are applied after any de-quantization.", "", ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (dequant, bool, "Dequantize output tensors to float32 from their native quantized type", true, ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (tpunum, size_t, "Coral EdgeTPU number to use to run this model, typically 0, or can be " "1 when using a dual-TPU add-on board, or more when using additional TPUs connected " "to USB ports", 0, ParamCateg)
	Parameter.
	JEVOIS_DECLARE_PARAMETER (spunum, size_t, "Hailo8 device number to use to run this model, typically 0 unless " "several Hailo8 accelerators are connected to the system", 0, ParamCateg)
	Parameter.

Constructor & Destructor Documentation

~Network()

jevois::dnn::Network::~Network ( )

virtual

Destructor.

CAUTION: derived classes must call waitBeforeDestroy() in their destructor

Definition at line 28 of file Network.C.

Member Function Documentation

doprocess()

virtual std::vector< cv::Mat > jevois::dnn::Network::doprocess ( std::vector< cv::Mat > const &  blobs, std::vector< std::string > &  info  )

protectedpure virtual

Process input blobs and obtain output blobs.

Implemented in jevois::dnn::NetworkHailo, jevois::dnn::NetworkNPU, jevois::dnn::NetworkONNX, jevois::dnn::NetworkOpenCV, jevois::dnn::NetworkPython, and jevois::dnn::NetworkTPU.

freeze()

void jevois::dnn::Network::freeze ( bool  doit )

virtual

Freeze/unfreeze parameters that users should not change while running.

Note: derived classes can freeze their own params by overriding this function, and should remember to still call the base class jevois::dnn::Network::freeze(doit)

Reimplemented in jevois::dnn::NetworkHailo, jevois::dnn::NetworkNPU, jevois::dnn::NetworkONNX, jevois::dnn::NetworkOpenCV, jevois::dnn::NetworkPython, and jevois::dnn::NetworkTPU.

Definition at line 32 of file Network.C.

Referenced by jevois::dnn::NetworkHailo::freeze(), jevois::dnn::NetworkNPU::freeze(), jevois::dnn::NetworkONNX::freeze(), jevois::dnn::NetworkOpenCV::freeze(), jevois::dnn::NetworkPython::freeze(), and jevois::dnn::NetworkTPU::freeze().

inputShapes()

virtual std::vector< vsi_nn_tensor_attr_t > jevois::dnn::Network::inputShapes ( )

pure virtual

Get shapes of all input tensors.

Implemented in jevois::dnn::NetworkHailo, jevois::dnn::NetworkNPU, jevois::dnn::NetworkONNX, jevois::dnn::NetworkOpenCV, jevois::dnn::NetworkPython, and jevois::dnn::NetworkTPU.

load()

virtual void jevois::dnn::Network::load ( )

protectedpure virtual

Load from disk.

Implemented in jevois::dnn::NetworkHailo, jevois::dnn::NetworkNPU, jevois::dnn::NetworkONNX, jevois::dnn::NetworkOpenCV, jevois::dnn::NetworkPython, and jevois::dnn::NetworkTPU.

Referenced by ready().

onParamChange()

void jevois::dnn::Network::onParamChange ( network::outtransform const &  param, std::string const &  val  )

overrideprotected

Definition at line 40 of file Network.C.

References ALL_TENSORS, LFATAL, o, jevois::split(), and jevois::dnn::strshape().

outputShapes()

virtual std::vector< vsi_nn_tensor_attr_t > jevois::dnn::Network::outputShapes ( )

pure virtual

Get shapes of all output tensors.

Implemented in jevois::dnn::NetworkHailo, jevois::dnn::NetworkNPU, jevois::dnn::NetworkONNX, jevois::dnn::NetworkOpenCV, jevois::dnn::NetworkPython, and jevois::dnn::NetworkTPU.

process()

std::vector< cv::Mat > jevois::dnn::Network::process	(	std::vector< cv::Mat > const &	blobs,
		std::vector< std::string > &	info
	)

Process input blobs and obtain output blobs.

Network implementations may push information data into the info string, which will be displayed to user. Convention is: if an info line starts with '* ', it is a header, and if it starts with '- ' it is a bullet. Info should always be organized into headers at the top level.

Definition at line 201 of file Network.C.

References ALL_TENSORS, jevois::async(), jevois::dnn::attrmat(), jevois::dnn::concatenate(), jevois::join(), jevois::joinall(), LFATAL, o, jevois::dnn::shapestr(), jevois::dnn::split(), jevois::split(), jevois::TimerOne::start(), jevois::TimerOne::stop(), and jevois::dnn::strshape().

ready()

bool jevois::dnn::Network::ready

(

)

Returns true when network is ready to run (loaded and initialized)

Definition at line 165 of file Network.C.

References jevois::async(), LINFO, load(), and jevois::warnAndRethrowException().

Referenced by jevois::dnn::NetworkNPU::inputShapes().

waitBeforeDestroy()

void jevois::dnn::Network::waitBeforeDestroy

(

)

If network is currently loading, wait until that is done before destroying.

CAUTION: derived classes must call waitBeforeDestroy() in their destructor

Definition at line 152 of file Network.C.

References LINFO.

Referenced by jevois::dnn::NetworkOpenCV::~NetworkOpenCV(), and jevois::dnn::NetworkTPU::~NetworkTPU().

Friends And Related Symbol Documentation

JEVOIS_DECLARE_PARAMETER() [1/13]

JEVOIS_DECLARE_PARAMETER ( backend  , Backend  , "OpenCV compute backend to use. Default will use the inference " "engine if  available, otherwise OpenCV(note that inference engine only works on Intel " "processors or MyriadX hardware, thus you should normally select OpenCV when running " "on JeVois-Pro Platform, unless you want to use an optional MyriadX accelerator). " "Changes will take effect next time you load a model."  , JEVOIS_BACKEND_DEFAULT  , Backend_Values  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [2/13]

JEVOIS_DECLARE_PARAMETER ( comment  , std::string  , "Optional comment about the network"  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [3/13]

JEVOIS_DECLARE_PARAMETER ( config  , std::string  , "Path to a text file that contains network configuration. " "Can have extension .prototxt   Caffe, .  pbtxtTensorFlow, or .cfg(Darknet). " "If path is  relative, it will be prefixed by dataroot."  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [4/13]

JEVOIS_DECLARE_PARAMETER ( dataroot  , std::string  , "Root directory to use when config or model parameters " "are relative paths."  , JEVOIS_SHARE_PATH  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [5/13]

JEVOIS_DECLARE_PARAMETER ( dequant  , bool  , "Dequantize output tensors to float32 from their native quantized type"  , true  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [6/13]

JEVOIS_DECLARE_PARAMETER ( extraintensors  , std::string  , "Specification of extra fixed input tensors that will be " "added after the regular intensors. Format is: " "<type>:<shape>:val1 val2 ...  valN, < type >:< shape >:val1 ... valN. For  example, for " "URetinex-Net:32F:1x1x1:3.0"  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [7/13]

JEVOIS_DECLARE_PARAMETER ( intensors  , std::string  , "Specification of input tensors"  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [8/13]

JEVOIS_DECLARE_PARAMETER ( model  , std::string  , "Path to a binary file of model contains trained weights. " "Can have extension .caffemodel   Caffe, .  pbTensorFlow, .t7 or .  netTorch, " ".  tfliteTensorFlow Lite, or .weights(Darknet). If path is  relative, it will be " "prefixed by dataroot."  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [9/13]

JEVOIS_DECLARE_PARAMETER ( outtensors  , std::string  , "Specification of output tensors"  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [10/13]

JEVOIS_DECLARE_PARAMETER ( spunum  , size_t  , "Hailo8 device number to use to run this  model, typically 0 unless " "several Hailo8 accelerators are connected to the system"  , 0  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [11/13]

JEVOIS_DECLARE_PARAMETER ( target  , Target  , "OpenCV compute target to use. Changes will take effect " "next time you load a different model."  , Target::CPU  , Target_Values  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [12/13]

JEVOIS_DECLARE_PARAMETER ( tpunum  , size_t  , "Coral EdgeTPU number to use to run this  model, typically  0, or can be " "1 when using a dual-TPU add-on  board, or more when using additional TPUs connected " "to USB ports"  , 0  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER() [13/13]

JEVOIS_DECLARE_PARAMETER ( url  , std::string  , "Optional URL for the network"  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DECLARE_PARAMETER_WITH_CALLBACK()

JEVOIS_DECLARE_PARAMETER_WITH_CALLBACK ( outtransform  , std::string  , "Specification of a sequence of operations " "(reshape, transpose, split, merge, etc) to be applied to the output " "tensors. Useful to match a particular network's outputs to the " "tensor shapes expected by a particular post-processor. The sequence of " "operations is separated by  semicolons, and applied from left to right. " "Operations specified here are applied after any de-quantization."  , ""  , ParamCateg    )

related

Parameter.

JEVOIS_DEFINE_ENUM_CLASS() [1/2]

JEVOIS_DEFINE_ENUM_CLASS ( Backend  , (OpenCV)(InferenceEngine)(TimVX)    )

related

Enum.

JEVOIS_DEFINE_ENUM_CLASS() [2/2]

JEVOIS_DEFINE_ENUM_CLASS ( Target  , (CPU)(OpenCL)(OpenCL_FP16)(Myriad)(NPU)    )

related

Enum.

---

The documentation for this class was generated from the following files:

• include/jevois/DNN/Network.H
• src/jevois/DNN/Network.C