The DetectionOutput plugin layer generates the detection output based on location and confidence predictions by doing non maximum suppression. More...

#include <NvInferPluginUtils.h>

Public Attributes
bool	shareLocation
	If true, bounding box are shared among different classes. More...

bool	varianceEncodedInTarget

int32_t	backgroundLabelId
	Background label ID. If there is no background class, set it as -1. More...

int32_t	numClasses
	Number of classes to be predicted. More...

int32_t	topK

int32_t	keepTopK
	Number of total bounding boxes to be kept per image after NMS step. More...

float	confidenceThreshold
	Only consider detections whose confidences are larger than a threshold. More...

float	nmsThreshold
	Threshold to be used in NMS. More...

CodeTypeSSD	codeType
	Type of coding method for bbox. More...

int32_t	inputOrder [3]
	Specifies the order of inputs {loc_data, conf_data, priorbox_data}. More...

bool	confSigmoid
	Set to true to calculate sigmoid of confidence scores. More...

bool	isNormalized
	Set to true if bounding box data is normalized by the network. More...

bool	isBatchAgnostic {true}
	Defaults to true. Set to false if prior boxes are unique per batch. More...

Detailed Description

The DetectionOutput plugin layer generates the detection output based on location and confidence predictions by doing non maximum suppression.

This plugin first decodes the bounding boxes based on the anchors generated. It then performs non_max_suppression on the decoded bounding boxes. DetectionOutputParameters defines a set of parameters for creating the DetectionOutput plugin layer.

Deprecated:: Deprecated in TensorRT 10.0. DetectionOutput plugin is deprecated.

Member Data Documentation

◆ backgroundLabelId

int32_t nvinfer1::plugin::DetectionOutputParameters::backgroundLabelId

Background label ID. If there is no background class, set it as -1.

◆ codeType

CodeTypeSSD nvinfer1::plugin::DetectionOutputParameters::codeType

Type of coding method for bbox.

◆ confidenceThreshold

float nvinfer1::plugin::DetectionOutputParameters::confidenceThreshold

Only consider detections whose confidences are larger than a threshold.

◆ confSigmoid

bool nvinfer1::plugin::DetectionOutputParameters::confSigmoid

Set to true to calculate sigmoid of confidence scores.

◆ inputOrder

int32_t nvinfer1::plugin::DetectionOutputParameters::inputOrder[3]

Specifies the order of inputs {loc_data, conf_data, priorbox_data}.

◆ isBatchAgnostic

bool nvinfer1::plugin::DetectionOutputParameters::isBatchAgnostic {true}

Defaults to true. Set to false if prior boxes are unique per batch.

◆ isNormalized

bool nvinfer1::plugin::DetectionOutputParameters::isNormalized

Set to true if bounding box data is normalized by the network.

◆ keepTopK

int32_t nvinfer1::plugin::DetectionOutputParameters::keepTopK

Number of total bounding boxes to be kept per image after NMS step.

◆ nmsThreshold

float nvinfer1::plugin::DetectionOutputParameters::nmsThreshold

Threshold to be used in NMS.

◆ numClasses

int32_t nvinfer1::plugin::DetectionOutputParameters::numClasses

Number of classes to be predicted.

◆ shareLocation

bool nvinfer1::plugin::DetectionOutputParameters::shareLocation

If true, bounding box are shared among different classes.

◆ topK

int32_t nvinfer1::plugin::DetectionOutputParameters::topK

Number of boxes per image with top confidence scores that are fed into the NMS algorithm.

◆ varianceEncodedInTarget

bool nvinfer1::plugin::DetectionOutputParameters::varianceEncodedInTarget

If true, variance is encoded in target. Otherwise we need to adjust the predicted offset accordingly.

The documentation for this struct was generated from the following file:

NvInferPluginUtils.h

Public Attributes