ds::Settings Class Reference

DeepSight SDK settings class. More...

Public Types
enum	OperationMode : int { OperationMode::DEVELOPMENT, OperationMode::RELEASE_LIFETIME, OperationMode::RELEASE_SESSION }
	The mode the SDK should operate when authenticating. More...
enum	LogLevel : int {   LogLevel::TRACE, LogLevel::DEBUG, LogLevel::INFO, LogLevel::WARN,   LogLevel::ERR, LogLevel::CRITICAL }
	The log-level that should internally be used by the SDK. More...
enum	DnnTarget : int { DnnTarget::CPU, DnnTarget::GPU, DnnTarget::GPU_MT }
	Target processor on which the SDK's neural networks should execute. More...
enum	TrackerFps : int { TrackerFps::DYNAMIC, TrackerFps::FPS_30, TrackerFps::FPS_10, TrackerFps::FPS_5 }
	Frame processing speed for which the face counting or people counting algorithm should be tuned. More...
enum	Range : int { Range::REGULAR, Range::LONG }
	Selector for switching the type of Face Detector to use. More...
enum	NetworkType : int { NetworkType::NORMAL, NetworkType::QUANTIZED }
	Selector for the type of network to use when performing analysis. The SDK offers two types of networks to use, normal or quantized networks. The normal networks are used for the best accuracy while the quantized networks are optimized faster execution. More...

Public Member Functions
void	load (const std::string &filePath)
	Loads an ini settings file into the ds::Settings instance. More...
void	save (const std::string &filePath)
	Saves the object's parameters into a DeepSight settings ini file. More...
bool	operator!= (const Settings &settings) const
	Returns true if the this object is not equal to settings. Inequality is tested member-wise.
bool	operator== (const Settings &settings) const
	Returns true if the this object is equal to settings. Equality is tested member-wise.
bool	cpuTargetEnabled () const
	Returns true if the dnnTarget value for this settings object is set to 'CPU'.
bool	gpuTargetEnabled () const
	Returns true if the dnnTarget value for this settings object is set to 'GPU' or 'GPU_MT'.

Public Attributes
std::string	neuralNetworksPath = ""
	Path of the directory containing the SDK neural network files. More...
std::string	licenseKey = ""
	License key used for authenticating the SDK.
std::string	lifetimeAuthFile = ""
	Path to file where the authorization for the lifetime release license is saved. More...
OperationMode	operationMode = OperationMode::DEVELOPMENT
	Operation mode the SDK should authenticate with.
LogLevel	logLevel = LogLevel::INFO
	Log-level the SDK should operate with. Accepted enum values are 'INFO', 'WARN', 'ERR', and 'CRITICAL'. 'TRACE' and 'DEBUG' are reserved for internal use.
DnnTarget	dnnTarget = DnnTarget::CPU
	Target processor on which the SDK's neural networks should execute. Accepted enum values are 'CPU', 'GPU', and 'GPU_MT'. More...
TrackerFps	bodyTrackFps = TrackerFps::DYNAMIC
	Tunes the people counting algorithm to the video input fps. Use ds::Settings::DYNAMIC for live feed input cameras and use ds::Settings::FPS_30 for video files (most commonly encoded at 30 fps).
TrackerFps	faceTrackFps = TrackerFps::DYNAMIC
	Tunes the face counting algorithm to the video input fps. Use ds::Settings::DYNAMIC for live feed input cameras and use ds::Settings::FPS_30 for video files (most commonly encoded at 30 fps).
NetworkType	networkType = NetworkType::NORMAL
	Selects the type of networks that the SDK should use. Accepted enum values are 'NORMAL' and 'QUANTIZED'.
bool	useAge = true
	Tells the SDK whether it should perform age estimation for each detected face.
bool	useGender = true
	Tells the SDK whether it should perform gender estimation for each detected face.
bool	useHeadpose = true
	Tells the SDK whether it should perform head pose estimation for each detected face.
bool	useSmile = true
	Tells the SDK whether it should perform smile estimation for each detected face.
bool	useFaceMask = true
	Tells the SDK whether it should perform face mask estimation for each detected face.
bool	useEmotions = true
	Tells the SDK whether it should perform emotion estimation for each detected face.
float	faceConfidenceThreshold = 0.8f
	Face confidence threshold. More...
int	maxNumFaces = -1
	Controls the maximum number of faces that will be returned per frame after detection. A value of -1 (default) indicates no limit.
int	minFaceSize = 0
	Unless set to 0, faces smaller than this value will be discarded after detection.
int	maxFaceSize = 0
	Unless set to 0, faces bigger than this value will be discarded after detection.
Range	detectionRange = Range::REGULAR
	Selects the face detector to be used. More...
int	numThreads = -1
	Controls the number of threads used to execute parallel computation regions, like inference. More...

Detailed Description

DeepSight SDK settings class.

The ds::Settings class is a container for configuration parameters used during runtime. By changing the settings values it is possible to control the SDK behavior.

Member Enumeration Documentation

DnnTarget

enum ds::Settings::DnnTarget : int

strong

Target processor on which the SDK's neural networks should execute.

Enumerator
CPU	Uses the CPU for performing inference on the networks.
GPU	Uses the GPU for accelerating the inference on the networks. Refer to the OpenCL acceleration OpenCL section.
GPU_MT	Uses the GPU for accelerating the inference performed on the SDK's networks. Inference of the different networks will be performed in parallel. Refer to the OpenCL acceleration OpenCL section.

LogLevel

enum ds::Settings::LogLevel : int

strong

The log-level that should internally be used by the SDK.

Enumerator
TRACE	Trace logging level (reserved for internal use).
DEBUG	Debug logging level (reserved for internal use).
INFO	Info logging level.
WARN	Warning logging level.
ERR	Error logging level.
CRITICAL	Critical logging level.

NetworkType

enum ds::Settings::NetworkType : int

strong

Selector for the type of network to use when performing analysis. The SDK offers two types of networks to use, normal or quantized networks. The normal networks are used for the best accuracy while the quantized networks are optimized faster execution.

Enumerator
NORMAL	Use the default networks for optimal accuracy.
QUANTIZED	Use the quantized networks for higher performance.

OperationMode

enum ds::Settings::OperationMode : int

strong

The mode the SDK should operate when authenticating.

The SDK license can operate in different modalities. This enumerations holds the modalities the SDK can use the license with.

Enumerator
DEVELOPMENT	Uses the license in development mode.
RELEASE_LIFETIME	Uses the license in lifetime release mode.
RELEASE_SESSION	Uses the license in session release mode.

Range

enum ds::Settings::Range : int

strong

Selector for switching the type of Face Detector to use.

Enumerator
REGULAR	Use the regular range Face Detector. This detector is useful if you want to detect people on a short to medium range and save on computation resources. This is the default Face Detector and detects at the same range compared to previous versions of the SDK.
LONG	Use the long range Face Detector. This detector is useful if you want to detect people on a medium to long range and is computationally more expensive.

TrackerFps

enum ds::Settings::TrackerFps : int

strong

Frame processing speed for which the face counting or people counting algorithm should be tuned.

Enumerator
DYNAMIC	Dynamically tunes the tracker to the current fps speed. To be used with real-time input like webcams of IP cameras.
FPS_30	Tunes the tracker to an input speed of 30 fps. To be used only with video files encoded at about 30 fps (most common encoding).
FPS_10	Tunes the tracker to an input speed of 10 fps. To be used only with video files encoded at about 10 fps (almost never used).
FPS_5	Tunes the tracker to an input speed of 5 fps. To be used only with video files encoded at about 5 fps (almost never used).

Member Function Documentation

load()

void ds::Settings::load

(

const std::string &

filePath

)

Loads an ini settings file into the ds::Settings instance.

Parameters

filePath

Path of a file storing DeepSight settings.

Throws an std::runtime_error if the file cannot be loaded.

Example usage:

ds::Settings s;
try
{
  s.load( "/path/to/settings" )
}
catch( const std::runtime_error & e )
{
  std::cerr << "Could not load settings : " << e.what() << std::endl;
}

save()

void ds::Settings::save

(

const std::string &

filePath

)

Saves the object's parameters into a DeepSight settings ini file.

Parameters

filePath

Path of a file where to store DeepSight settings.

Throws an std::runtime_error if the file cannot be saved.

Example usage:

ds::Settings s;
try
{
  s.save( "/path/to/settings" )
}
catch( const std::runtime_error & e )
{
  std::cerr << "Could not save settings : " << e.what() << std::endl;
}

Member Data Documentation

detectionRange

ds::Settings::Range ds::Settings::detectionRange = Range::REGULAR

Selects the face detector to be used.

Two face detector neural networks are available as part of the SDK. One is optimized for long range detection, the other for medium range.

Note

The long rage face detector is slower than the regular one.

Faces detected on longer distances are smaller than the faces close by. Accuracy of features (eg. age, gender, headpose and so on) on smaller faces is lower.

dnnTarget

ds::Settings::DnnTarget ds::Settings::dnnTarget = DnnTarget::CPU

Target processor on which the SDK's neural networks should execute. Accepted enum values are 'CPU', 'GPU', and 'GPU_MT'.

To enable GPU acceleration refer to OpenCL acceleration.

Note

The GPU_MT option is currently experimental and enables concurrent GPU network inference. This may cause a gradual loss of performance over time.

faceConfidenceThreshold

float ds::Settings::faceConfidenceThreshold = 0.8f

Face confidence threshold.

This threshold controls the confidence level the SDK returns faces with. Faces detected during ds::DeepSight::analyze and ds::DeepSight::analyzeSingleImage with a confidence lower than the specified faceConfidenceThreshold will be discarded. Valid values are in the range [0, 1].

lifetimeAuthFile

std::string ds::Settings::lifetimeAuthFile = ""

Path to file where the authorization for the lifetime release license is saved.

Use in combination with ds::Settings::operationMode set to ds::Settings::OperationMode::RELEASE_LIFETIME.

Example:

ds::Settings s;
s.operationMode = ds::Settings::OperationMode::RELEASE_LIFETIME;
s.lifetimeAuthFile = "/path/to/lifetime_authorization.bin";
ds::DeepSight ds{ settings };
ds.authenticate();

neuralNetworksPath

std::string ds::Settings::neuralNetworksPath = ""

Path of the directory containing the SDK neural network files.

Failing to properly set this parameter causes the ds::DeepSight constructor to fail and throw an exception.

numThreads

int ds::Settings::numThreads = -1

Controls the number of threads used to execute parallel computation regions, like inference.

This parameter controls how much CPU is used on multicore systems. Set to -1 to use the system default. Setting this parameter to a value lower than the number of CPU cores will leave the remaining CPU cores available for other computation. Eg. On a 4-core setup, setting ds::Settings::numThreads to 2 will leave ~50% of CPU available for other processes.

Note

This parameter is set internally using cv::setNumThreads() at the beginning of each ds::DeepSight::analyze or ds::DeepSight::analyzeSingleImage call. If cv::setNumThreads() is used externally, the number of threads will be set to ds::Settings::numThreads at the next call of ds::DeepSight::analyze or ds::DeepSight::analyzeSingleImage.