Classes
class	AI_MODELS
	Lists available AI models. More...

class	OBJECT_DETECTION_MODEL
	Lists available models for the object detection module. More...

class	OBJECT_FILTERING_MODE
	Lists supported bounding box preprocessing. More...

class	OBJECT_ACCELERATION_PRESET
	Lists supported presets for maximum acceleration allowed for a given tracked object. More...

class	OBJECT_CLASS
	Lists available object classes. More...

class	OBJECT_SUBCLASS
	List available object subclasses. More...

class	OBJECT_TRACKING_STATE
	Lists the different states of object tracking. More...

class	OBJECT_ACTION_STATE
	Lists the different states of an object's actions. More...

class	ObjectData
	Class containing data of a detected object such as its bounding_box, label, id and its 3D position. More...

class	CustomBoxObjectData
	Class that store externally detected objects. More...

class	CustomMaskObjectData
	Class storing externally detected objects. More...

class	ObjectsBatch
	Class containing batched data of a detected objects from the object detection module. More...

class	Objects
	Class containing the results of the object detection module. More...

class	BatchParameters
	Class containing a set of parameters for batch object detection. More...

class	ObjectDetectionParameters
	Class containing a set of parameters for the object detection module. More...

class	ObjectTrackingParameters
	Class containing tracking parameters for object detection. More...

class	ObjectDetectionRuntimeParameters
	Class containing a set of runtime parameters for the object detection module. More...

class	ObjectTrackingParametersProxy
	Class containing a set of runtime properties of a certain class ID for the object detection module using a custom model. More...

class	CustomObjectDetectionRuntimeParameters
	Class containing a set of runtime parameters for the object detection module using your own model ran by the SDK. More...

Functions
def	generate_unique_id ()
	Generate a UUID like unique id to help identify and track AI detections.

def	optimize_custom_ai_model (str custom_onnx_file, Resolution custom_onnx_dynamic_input_shape=Resolution(512, 512), int, gpu_id=0)
	Optimizes a custom object detection model ahead of time, so that Camera.enable_object_detection() can start using it right away. More...

Function Documentation

◆ optimize_custom_ai_model()

def pyzed.sl.optimize_custom_ai_model	(	str	custom_onnx_file,
		Resolution	custom_onnx_dynamic_input_shape = `Resolution(512, 512)`,
			int,
			gpu_id = `0`
	)

Optimizes a custom object detection model ahead of time, so that Camera.enable_object_detection() can start using it right away.

This optimizes the given ONNX file exactly as Camera.enable_object_detection() does when ObjectDetectionParameters.detection_model is set to OBJECT_DETECTION_MODEL.CUSTOM_YOLOLIKE_BOX_OBJECTS, OBJECT_DETECTION_MODEL.CUSTOM_RFDETRLIKE_BOX_OBJECTS or OBJECT_DETECTION_MODEL.CUSTOM_BOX_OBJECTS_AUTODETECT, and saves the result for re-use. Optimizing a model can take several minutes, so this is meant to be called once when installing or deploying your application, rather than on its critical path.

Parameters

custom_onnx_file	: Path to the ONNX file to optimize. Use the same value as ObjectDetectionParameters.custom_onnx_file.
custom_onnx_dynamic_input_shape	: Input resolution to optimize the model for. Use the same value as ObjectDetectionParameters.custom_onnx_dynamic_input_shape, otherwise the optimized model cannot be re-used and the model is optimized again at runtime.
gpu_id	: ID of the GPU on which the model will run. The optimized model is specific to it.

Returns: sl.ERROR_CODE.SUCCESS if the model is optimized and ready to be used.; sl.ERROR_CODE.INVALID_FUNCTION_PARAMETERS if no ONNX file was given, or if it cannot be found.

Note: A model with a fixed input resolution keeps its own: custom_onnx_dynamic_input_shape is then only used to identify the optimized model.

Referenced by InitParametersOne.set_from_holoscan_port().

Classes

Functions

Function Documentation

◆ optimize_custom_ai_model()