#include <nms_cuda.h>

apollo::perception::lidar::NmsCuda 的协作图:

Public 成员函数
	NmsCuda (const int num_threads, const int num_box_corners, const float nms_overlap_threshold)
	Constructor

void	DoNmsCuda (const int host_filter_count, float dev_sorted_box_for_nms, int out_keep_inds, int *out_num_to_keep)
	GPU Non-Maximum Suppresion for network output

详细描述

在文件 nms_cuda.h 第 53 行定义.

构造及析构函数说明

apollo::perception::lidar::NmsCuda::NmsCuda	(	const int	num_threads,
		const int	num_box_corners,
		const float	nms_overlap_threshold
	)

Constructor

参数

[in]	num_threads	Number of threads when launching cuda kernel
[in]	num_box_corners	Number of corners for 2D box
[in]	nms_overlap_threshold	IOU threshold for NMS

Captital variables never change after the compile, Non-captital variables could be chaned through rosparam

void apollo::perception::lidar::NmsCuda::DoNmsCuda	(	const int	host_filter_count,
		float *	dev_sorted_box_for_nms,
		int *	out_keep_inds,
		int *	out_num_to_keep
	)

GPU Non-Maximum Suppresion for network output

参数

[in]	host_filter_count	Number of filtered output
[in]	dev_sorted_box_for_nms	Bounding box output sorted by score
[out]	out_keep_inds	Indexes of selected bounding box
[out]	out_num_to_keep	Number of kept bounding boxes

NMS in GPU and postprocessing for selecting box in CPU

该类的文档由以下文件生成: