apollo::perception::inference Namespace Reference

Classes
class	ArgMax1Plugin
class	BatchStream
class	CaffeNet
struct	ConvParam
class	CudaUtil
class	DFMBPSROIAlignPlugin
class	GPUL2Norm
class	Inference
class	Layer
class	Logger
class	ObstacleDetector
class	OnnxObstacleDetector
class	RCNNProposalPlugin
class	ReLUPlugin
class	ROIPoolingLayer
class	RPNProposalSSDPlugin
class	RTNet
class	SLICEPlugin
class	SoftmaxPlugin
class	TorchDet
class	TorchNet

Typedefs
using	BlobPtr = std::shared_ptr< apollo::perception::base::Blob< float > >
typedef std::map< std::string, std::shared_ptr< apollo::perception::base::Blob< float > > >	BlobMap
typedef std::map< std::string, std::vector< nvinfer1::Weights > >	WeightMap
typedef std::map< std::string, nvinfer1::ITensor * >	TensorMap
typedef std::map< std::string, nvinfer1::DimsCHW >	TensorDimsMap
typedef std::map< std::string, std::string >	TensorModifyMap

Functions
Inference *	CreateInferenceByName (const std::string &name, const std::string &proto_file, const std::string &weight_file, const std::vector< std::string > &outputs, const std::vector< std::string > &inputs, const std::string &model_root="")
void	bbox_transform_inv_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, const float *boxes, const float *deltas, const int num_box, const int num_channel, float *out_boxes)
void	clip_boxes_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, float *boxes, const float height, const float width)
void	filter_boxes_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, const float *boxes, const float *scores, const float *all_probs, const int num_box, const int num_channel, const int num_class, const int num_prob, const int filter_channel, const int filter_class, const int min_size_mode, const float min_size_h, const float min_size_w, const float threshold_score, float *filtered_boxes, float *filtered_scores, float *filtered_all_probs, int *filtered_count)
void	keep_topN_boxes_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, const float *boxes, const float *scores, const float *all_probs, const int *indexes, const int *count, const bool keep_score, const int num_box, const int num_prob, const int topN, float *out_boxes, float *out_scores, float *out_all_probs)
void	repeatedly_add_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, const float *in_data, float *out_data, const float *add_vec, int add_vec_size)
void	repeatedly_mul_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, const float *in_data, float *out_data, const float *mul_vec, int mul_vec_size)
void	slice2d_cuda (int block_size, int thread_size, int shared_mem, cudaStream_t stream, const int nthreads, const float *in_data, float *out_data, const int *slice_axises, int slice_axis_num, int input_axis_size)
void	NmsForward (bool rpn_proposal_output_score, int host_filter_count, int num_box_corners, float nms_overlap_threshold, int num_candidate, int top_n, int batch_id, int num_prob, float *dev_sorted_box_for_nms, float *scores, float *all_probs, float *out_boxes, int *acc_box_num, cudaStream_t stream)
	GPU Non-Maximum Suppresion for network output. More...
nvinfer1::DimsCHW	ReshapeDims (const nvinfer1::DimsCHW &dims, const nvinfer1::DimsCHW &inputDims)
void	ParseNetParam (const NetParameter &net_param, TensorDimsMap *tensor_dims_map, std::map< std::string, std::string > *tensor_modify_map, std::vector< LayerParameter > *order)
bool	modify_pool_param (PoolingParameter *pool_param)
bool	ParserConvParam (const ConvolutionParameter &conv, ConvParam *param)
nvinfer1::DimsCHW	getCHW (const nvinfer1::Dims &d)
bool	ReadProtoFromTextFile (const std::string &filename, google::protobuf::Message *proto)
bool	ReadProtoFromBinaryFile (const std::string &filename, google::protobuf::Message *proto)
bool	loadNetParams (const std::string &param_file, NetParameter *param)
std::string	locateFile (const std::string &path, const std::string &input)
size_t	BinaryReadString (FILE *fp, char *name)
size_t	BinaryWriteString (FILE *fp, const std::string &str)
template<typename Dtype >
std::shared_ptr< base::Blob< Dtype > >	BinaryReadBlob (FILE *fp)
template<typename Dtype >
void	BinaryWriteBlob (FILE *fp, const base::Blob< Dtype > &blob)
template<typename Dtype >
std::map< std::string, std::shared_ptr< base::Blob< Dtype > > >	BinaryReadFile (const char *file_path)
template<typename Btype >
bool	BinaryWriteFile (const char *file_path, const std::map< std::string, Btype > &data_dict)
void	GPUGemmFloat (const CBLAS_TRANSPOSE TransA, const CBLAS_TRANSPOSE TransB, const int M, const int N, const int K, const float alpha, const float *A, const float *B, const float beta, float *C)
void	GPUMultiFloat (const int n, const float *a, const float *b, float *result)
void	GPUMSetFloat (const int n, const float alpha, float *result)
bool	ResizeGPU (const base::Image8U &src, std::shared_ptr< apollo::perception::base::Blob< float >> dst, int stepwidth, int start_axis)
bool	ResizeGPU (const apollo::perception::base::Blob< uint8_t > &src_gpu, std::shared_ptr< apollo::perception::base::Blob< float >> dst, int stepwidth, int start_axis, int mean_b, int mean_g, int mean_r, bool channel_axis, float scale)
bool	ResizeGPU (const base::Image8U &src, std::shared_ptr< apollo::perception::base::Blob< float >> dst, int stepwidth, int start_axis, float mean_b, float mean_g, float mean_r, bool channel_axis, float scale)
template<typename T >
void	load_data (const std::string &filename, std::vector< T > *outputs)
std::shared_ptr< float >	load_binary_data (const std::string &filename)
bool	write_result (const std::string &out_path, const std::vector< float > &results)
bool	write_result (const std::string &out_path, const std::map< std::string, std::vector< float >> &results)

Variables
const std::map< EltwiseParameter::EltwiseOp, nvinfer1::ElementWiseOperation >	eltwise_map
const std::map< std::string, nvinfer1::ActivationType >	active_map
const std::vector< std::string >	_gpu_checklist
constexpr size_t	kMaxStrLen = 64
constexpr int	kMinDim = 1
constexpr int	kMaxDim = std::numeric_limits<int>::max()

Typedef Documentation

BlobMap

typedef std::map<std::string, std::shared_ptr<apollo::perception::base::Blob<float> > > apollo::perception::inference::BlobMap

BlobPtr

typedef std::shared_ptr< apollo::perception::base::Blob< float > > apollo::perception::inference::BlobPtr

TensorDimsMap

typedef std::map< std::string, nvinfer1::DimsCHW > apollo::perception::inference::TensorDimsMap

TensorMap

typedef std::map< std::string, nvinfer1::ITensor * > apollo::perception::inference::TensorMap

TensorModifyMap

typedef std::map<std::string, std::string> apollo::perception::inference::TensorModifyMap

WeightMap

typedef std::map< std::string, std::vector< nvinfer1::Weights > > apollo::perception::inference::WeightMap

Function Documentation

bbox_transform_inv_cuda()

void apollo::perception::inference::bbox_transform_inv_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		const float *	boxes,
		const float *	deltas,
		const int	num_box,
		const int	num_channel,
		float *	out_boxes
	)

BinaryReadBlob()

template<typename Dtype >

std::shared_ptr<base::Blob<Dtype> > apollo::perception::inference::BinaryReadBlob

(

FILE *

fp

)

BinaryReadFile()

template<typename Dtype >

std::map<std::string, std::shared_ptr<base::Blob<Dtype> > > apollo::perception::inference::BinaryReadFile

(

const char *

file_path

)

BinaryReadString()

size_t apollo::perception::inference::BinaryReadString	(	FILE *	fp,
		char *	name
	)

BinaryWriteBlob()

template<typename Dtype >

void apollo::perception::inference::BinaryWriteBlob	(	FILE *	fp,
		const base::Blob< Dtype > &	blob
	)

BinaryWriteFile()

template<typename Btype >

bool apollo::perception::inference::BinaryWriteFile	(	const char *	file_path,
		const std::map< std::string, Btype > &	data_dict
	)

BinaryWriteString()

size_t apollo::perception::inference::BinaryWriteString	(	FILE *	fp,
		const std::string &	str
	)

clip_boxes_cuda()

void apollo::perception::inference::clip_boxes_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		float *	boxes,
		const float	height,
		const float	width
	)

CreateInferenceByName()

Inference* apollo::perception::inference::CreateInferenceByName	(	const std::string &	name,
		const std::string &	proto_file,
		const std::string &	weight_file,
		const std::vector< std::string > &	outputs,
		const std::vector< std::string > &	inputs,
		const std::string &	model_root = ""
	)

filter_boxes_cuda()

void apollo::perception::inference::filter_boxes_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		const float *	boxes,
		const float *	scores,
		const float *	all_probs,
		const int	num_box,
		const int	num_channel,
		const int	num_class,
		const int	num_prob,
		const int	filter_channel,
		const int	filter_class,
		const int	min_size_mode,
		const float	min_size_h,
		const float	min_size_w,
		const float	threshold_score,
		float *	filtered_boxes,
		float *	filtered_scores,
		float *	filtered_all_probs,
		int *	filtered_count
	)

getCHW()

nvinfer1::DimsCHW apollo::perception::inference::getCHW ( const nvinfer1::Dims &  d )

inline

GPUGemmFloat()

void apollo::perception::inference::GPUGemmFloat	(	const CBLAS_TRANSPOSE	TransA,
		const CBLAS_TRANSPOSE	TransB,
		const int	M,
		const int	N,
		const int	K,
		const float	alpha,
		const float *	A,
		const float *	B,
		const float	beta,
		float *	C
	)

GPUMSetFloat()

void apollo::perception::inference::GPUMSetFloat	(	const int	n,
		const float	alpha,
		float *	result
	)

GPUMultiFloat()

void apollo::perception::inference::GPUMultiFloat	(	const int	n,
		const float *	a,
		const float *	b,
		float *	result
	)

keep_topN_boxes_cuda()

void apollo::perception::inference::keep_topN_boxes_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		const float *	boxes,
		const float *	scores,
		const float *	all_probs,
		const int *	indexes,
		const int *	count,
		const bool	keep_score,
		const int	num_box,
		const int	num_prob,
		const int	topN,
		float *	out_boxes,
		float *	out_scores,
		float *	out_all_probs
	)

load_binary_data()

std::shared_ptr<float> apollo::perception::inference::load_binary_data

(

const std::string &

filename

)

load_data()

template<typename T >

void apollo::perception::inference::load_data	(	const std::string &	filename,
		std::vector< T > *	outputs
	)

loadNetParams()

bool apollo::perception::inference::loadNetParams	(	const std::string &	param_file,
		NetParameter *	param
	)

locateFile()

std::string apollo::perception::inference::locateFile	(	const std::string &	path,
		const std::string &	input
	)

modify_pool_param()

bool apollo::perception::inference::modify_pool_param

(

PoolingParameter *

pool_param

)

NmsForward()

void apollo::perception::inference::NmsForward	(	bool	rpn_proposal_output_score,
		int	host_filter_count,
		int	num_box_corners,
		float	nms_overlap_threshold,
		int	num_candidate,
		int	top_n,
		int	batch_id,
		int	num_prob,
		float *	dev_sorted_box_for_nms,
		float *	scores,
		float *	all_probs,
		float *	out_boxes,
		int *	acc_box_num,
		cudaStream_t	stream
	)

GPU Non-Maximum Suppresion for network output.

Parameters

[in]	rpn_proposal_output_score	Whether to output scores
[in]	host_filter_count	Number of filtered output
[in]	num_box_corners	Number of corners for 2D box
[in]	nms_overlap_threshold	IOU threshold for NMS
[in]	num_candidate	Pre-defined maximum number of candidates
[in]	top_n	Pre-defined maximum number of output boxes
[in]	batch_id	Id of current batch instance
[in]	num_prob	Number of probs
[in]	dev_sorted_box_for_nms	Bounding box output sorted by score
[in]	scores	Scores of boxes
[in]	all_probs	Probs of boxes for all classes and objectness
[in]	out_boxes	Output boxes
[in]	acc_box_num	Accumulated box num

NMS in GPU and postprocessing for selecting box in CPU

ParseNetParam()

void apollo::perception::inference::ParseNetParam	(	const NetParameter &	net_param,
		TensorDimsMap *	tensor_dims_map,
		std::map< std::string, std::string > *	tensor_modify_map,
		std::vector< LayerParameter > *	order
	)

ParserConvParam()

bool apollo::perception::inference::ParserConvParam	(	const ConvolutionParameter &	conv,
		ConvParam *	param
	)

ReadProtoFromBinaryFile()

bool apollo::perception::inference::ReadProtoFromBinaryFile	(	const std::string &	filename,
		google::protobuf::Message *	proto
	)

ReadProtoFromTextFile()

bool apollo::perception::inference::ReadProtoFromTextFile	(	const std::string &	filename,
		google::protobuf::Message *	proto
	)

repeatedly_add_cuda()

void apollo::perception::inference::repeatedly_add_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		const float *	in_data,
		float *	out_data,
		const float *	add_vec,
		int	add_vec_size
	)

repeatedly_mul_cuda()

void apollo::perception::inference::repeatedly_mul_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		const float *	in_data,
		float *	out_data,
		const float *	mul_vec,
		int	mul_vec_size
	)

ReshapeDims()

nvinfer1::DimsCHW apollo::perception::inference::ReshapeDims	(	const nvinfer1::DimsCHW &	dims,
		const nvinfer1::DimsCHW &	inputDims
	)

ResizeGPU() [1/3]

bool apollo::perception::inference::ResizeGPU	(	const base::Image8U &	src,
		std::shared_ptr< apollo::perception::base::Blob< float >>	dst,
		int	stepwidth,
		int	start_axis
	)

ResizeGPU() [2/3]

bool apollo::perception::inference::ResizeGPU	(	const apollo::perception::base::Blob< uint8_t > &	src_gpu,
		std::shared_ptr< apollo::perception::base::Blob< float >>	dst,
		int	stepwidth,
		int	start_axis,
		int	mean_b,
		int	mean_g,
		int	mean_r,
		bool	channel_axis,
		float	scale
	)

ResizeGPU() [3/3]

bool apollo::perception::inference::ResizeGPU	(	const base::Image8U &	src,
		std::shared_ptr< apollo::perception::base::Blob< float >>	dst,
		int	stepwidth,
		int	start_axis,
		float	mean_b,
		float	mean_g,
		float	mean_r,
		bool	channel_axis,
		float	scale
	)

slice2d_cuda()

void apollo::perception::inference::slice2d_cuda	(	int	block_size,
		int	thread_size,
		int	shared_mem,
		cudaStream_t	stream,
		const int	nthreads,
		const float *	in_data,
		float *	out_data,
		const int *	slice_axises,
		int	slice_axis_num,
		int	input_axis_size
	)

write_result() [1/2]

bool apollo::perception::inference::write_result	(	const std::string &	out_path,
		const std::vector< float > &	results
	)

write_result() [2/2]

bool apollo::perception::inference::write_result	(	const std::string &	out_path,
		const std::map< std::string, std::vector< float >> &	results
	)

Variable Documentation

_gpu_checklist

const std::vector<std::string> apollo::perception::inference::_gpu_checklist

Initial value:

{
    "GeForce GTX 1080",    "GeForce GTX 1080 Ti", "Tesla P4",
    "Tesla P40",           "GeForce GTX 1070",    "GeForce GTX 1060",
    "Tesla V100-SXM2-16GB"}

active_map

const std::map<std::string, nvinfer1::ActivationType> apollo::perception::inference::active_map

Initial value:

{
    {"Sigmoid", nvinfer1::ActivationType::kSIGMOID},
    {"TanH", nvinfer1::ActivationType::kTANH},
    {"ReLU", nvinfer1::ActivationType::kRELU}}

eltwise_map

const std::map<EltwiseParameter::EltwiseOp, nvinfer1::ElementWiseOperation> apollo::perception::inference::eltwise_map

Initial value:

{
        {EltwiseParameter_EltwiseOp_PROD,
         nvinfer1::ElementWiseOperation::kPROD},
        {EltwiseParameter_EltwiseOp_SUM, nvinfer1::ElementWiseOperation::kSUM},
        {EltwiseParameter_EltwiseOp_MAX, nvinfer1::ElementWiseOperation::kMAX}}

kMaxDim

constexpr int apollo::perception::inference::kMaxDim = std::numeric_limits<int>::max()

kMaxStrLen

constexpr size_t apollo::perception::inference::kMaxStrLen = 64

kMinDim

constexpr int apollo::perception::inference::kMinDim = 1