Include this file to get access to the Darknet/YOLO C API. More...

Include dependency graph for darknet.h:

This graph shows which files directly or indirectly include this file:

Classes
struct	DarknetBox
	Bounding box used with normalized coordinates (between 0.0 and 1.0). More...

struct	DarknetDetection
	Everything Darknet knows about a specific detection. More...

struct	DarknetImage
	The structure DarknetImage is used to store a normalized RGB Darknet image. More...

Typedefs
typedef struct DarknetBox	box
	Bounding box used with normalized coordinates (between 0.0 and 1.0).

typedef struct DarknetBox	DarknetBox
	Bounding box used with normalized coordinates (between 0.0 and 1.0).

typedef struct DarknetDetection	DarknetDetection
	Everything Darknet knows about a specific detection.

typedef struct DarknetImage	DarknetImage
	The structure DarknetImage is used to store a normalized RGB Darknet image.

typedef void *	DarknetNetworkPtr
	An opaque pointer to a Darknet::Network object, without needing to expose the internals of the network structure.

typedef struct DarknetDetection	detection
	Everything Darknet knows about a specific detection.

typedef struct DarknetImage	image
	Darknet-style image (vector of floats).

Enumerations
enum	NMS_KIND { DEFAULT_NMS , GREEDY_NMS , DIOU_NMS , CORNERS_NMS }

Functions
void	calculate_binary_weights (DarknetNetworkPtr ptr)
	This is part of the original `C` API.

void	copy_image_from_bytes (DarknetImage im, char *pdata)
	This is part of the original `C` API. Convert from data pointer to a Darknet image. Used by the Python API.

void	cuda_set_device (int n)
	This is part of the original `C` API. This function does nothing when Darknet was built to run on the CPU.

void	darknet_add_skipped_class (DarknetNetworkPtr ptr, const int class_to_skip)
	This is the `C` equivalent to Darknet::add_skipped_class().

void	darknet_clear_skipped_classes (DarknetNetworkPtr ptr)
	This is the `C` equivalent to Darknet::clear_skipped_classes().

void	darknet_del_skipped_class (DarknetNetworkPtr ptr, const int class_to_include)
	This is the `C` equivalent to Darknet::del_skipped_classes().

void	darknet_fix_out_of_bound_values (DarknetNetworkPtr ptr, const bool toggle)
	This is the `C` equivalent to Darknet::fix_out_of_bound_values().

void	darknet_free_neural_network (DarknetNetworkPtr *ptr)
	This is the `C` equivalent to Darknet::free_neural_network().

DarknetNetworkPtr	darknet_load_neural_network (const char const cfg_filename, const char const names_filename, const char *const weights_filename)
	This is the `C` equivalent to Darknet::load_neural_network().

void	darknet_network_dimensions (DarknetNetworkPtr ptr, int w, int h, int *c)
	This is the `C` equivalent to Darknet::network_dimensions().

void	darknet_set_detection_threshold (DarknetNetworkPtr ptr, float threshold)
	This is the `C` equivalent to Darknet::set_detection_threshold().

void	darknet_set_gpu_index (int idx)
	This is the `C` equivalent to Darknet::set_gpu_index().

void	darknet_set_non_maximal_suppression_threshold (DarknetNetworkPtr ptr, float threshold)
	This is the `C` equivalent to Darknet::set_non_maximal_suppression_threshold().

void	darknet_set_trace (const bool flag)
	This is the `C` equivalent to Darknet::set_trace().

void	darknet_set_verbose (const bool flag)
	This is the `C` equivalent to Darknet::set_verbose().

void	darknet_show_version_info ()
	This is the `C` equivalent to Darknet::show_version_info().

const char *	darknet_version_short ()
	Return the short version string, such as `"3.0.163"`.

const char *	darknet_version_string ()
	Return the full version string, such as `"v3.0-163-g56145bf"`.

void	diounms_sort (detection *dets, int total, int classes, float thresh, NMS_KIND nms_kind, float beta1)
	This is part of the original `C` API. Non Maxima Suppression.

void	do_nms_obj (detection *dets, int total, int classes, float thresh)
	This is part of the original `C` API. Non Maxima Suppression.

void	do_nms_sort (detection *dets, int total, int classes, float thresh)
	This is part of the original `C` API. Non Maxima Suppression.

void	free_detections (detection *dets, int n)
	This is part of the original `C` API. Do not use in new code.

void	free_image (image im)
	This is part of the original `C` API.

void	free_network_ptr (DarknetNetworkPtr ptr)
	This is part of the original `C` API.

void	free_ptrs (void **ptrs, int n)
	This is part of the original `C` API.

detection *	get_network_boxes (DarknetNetworkPtr ptr, int w, int h, float thresh, float hier, int map, int relative, int num, int letter)
	This is part of the original `C` API.

DarknetImage	load_image_v2 (const char *filename, int desired_width, int desired_height, int channels)
	This is part of the original `C` API. This function exists for the Python API.

DarknetImage	make_empty_image (int w, int h, int c)
	This is part of the original `C` API.

DarknetImage	make_image (int w, int h, int c)
	This is part of the original `C` API.

detection *	make_network_boxes (DarknetNetworkPtr ptr, float thresh, int *num)
	This is part of the original `C` API.

float *	network_predict_image (DarknetNetworkPtr ptr, const DarknetImage im)
	This is part of the original `C` API.

float *	network_predict_image_letterbox (DarknetNetworkPtr ptr, DarknetImage im)
	This is part of the original `C` API.

float *	network_predict_ptr (DarknetNetworkPtr ptr, float *input)
	This is part of the original `C` API.

void	reset_rnn (DarknetNetworkPtr ptr)
	This is part of the original `C` API.


DarknetNetworkPtr	load_network (const char cfg, const char weights, int clear)

DarknetNetworkPtr	load_network_custom (const char cfg, const char weights, int clear, int batch)

Detailed Description

Include this file to get access to the Darknet/YOLO C API.

The new (V3, July 2024) Darknet/YOLO C API is at the top of the file.
The original V2 Darknet/YOLO C API is at the bottom of this file (DARKNET_INCLUDE_ORIGINAL_API).
The new (V3, July 2024) Darknet/YOLO C++ API is in darknet.hpp.

https://darknetcv.ai/api/api.html

The old C V2 API did not have "darknet" in the function names nor the structures returned. It defined things like network and image in the global namespace, which can cause problems since those are common words. By default this old API is no longer exposed. If you're using some old software that expects the original C API in the darknet library, then make sure you "#define DARKNET_INCLUDE_ORIGINAL_API" before you include this header file.

Internally, Darknet still uses the old C V2 API.

Typedef Documentation

◆ box

typedef struct DarknetBox box

Bounding box used with normalized coordinates (between 0.0 and 1.0).

◆ DarknetBox

typedef struct DarknetBox DarknetBox

Bounding box used with normalized coordinates (between 0.0 and 1.0).

◆ DarknetDetection

typedef struct DarknetDetection DarknetDetection

Everything Darknet knows about a specific detection.

This structure is used by the old API. If using the new API, use Darknet::Prediction instead.

◆ DarknetImage

typedef struct DarknetImage DarknetImage

The structure DarknetImage is used to store a normalized RGB Darknet image.

The format is intended to be used for internal use by Darknet, but there are some situations where it may also be used or referenced externally via the Darknet API.

Moving forward starting with Darknet V3 (August 2024), where possible the emphasis will be to use OpenCV cv::Mat objects in the external-facing API instead of DarknetImage. Internally, Darknet will continue to use Image.

Warning: Keep this structure as POD (plain-old-data) since there are many places in the old code where memory for these image objects is calloc'd. This structure was originally part of the old Darknet C API, which is why it exists this way and not as a C++ class with methods.

Within data, the image is stored as 3 non-interlaced channels in RGB order. Each channel is stored as rows, from top to bottom. So a 5x3 RGB image would look like this:

r r r r r
r r r r r
r r r r r
g g g g g
g g g g g
g g g g g
b b b b b
b b b b b
b b b b b

A 5x3 pure red image, with a blueish/green square in the bottom-right corner would look like this:

0 1.0 1.0 1.0 1.0 // red channel
0 1.0 1.0 0.0 0.0 // note the lack of red in the bottom-right corner
0 1.0 1.0 0.0 0.0
0 0.0 0.0 0.0 0.0 // green channel
0 0.0 0.0 0.5 0.5
0 0.0 0.0 0.5 0.5
0 0.0 0.0 0.0 0.0 // blue channel
0 0.0 0.0 1.0 1.0 // note the blue square in the bottom-right corner
0 0.0 0.0 1.0 1.0

For additional information or to help debug the internals of DarknetImage, see Darknet::image_as_debug_string().

See also: Darknet::load_image(); Darknet::copy_image(); make_empty_image(); make_image(); Darknet::free_image()

◆ DarknetNetworkPtr

typedef void* DarknetNetworkPtr

An opaque pointer to a Darknet::Network object, without needing to expose the internals of the network structure.

◆ detection

typedef struct DarknetDetection detection

Everything Darknet knows about a specific detection.

◆ image

typedef struct DarknetImage image

Darknet-style image (vector of floats).

Enumeration Type Documentation

◆ NMS_KIND

enum NMS_KIND

See also: diounms_sort()

Enumerator
DEFAULT_NMS
GREEDY_NMS
DIOU_NMS
CORNERS_NMS

Function Documentation

◆ calculate_binary_weights()

void calculate_binary_weights ( DarknetNetworkPtr ptr )

This is part of the original C API.

Do not use in new code.

See also: darknet_load_neural_network(); Darknet::load_neural_network()

If you were previously using calculate_binary_weights() from within C code, it used to pass the network by value. Starting with Darknet V3 in 2024-08-16, the network is now passed in as a pointer.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ copy_image_from_bytes()

void copy_image_from_bytes	(	DarknetImage	im,
		char *	pdata
	)

This is part of the original C API. Convert from data pointer to a Darknet image. Used by the Python API.

◆ cuda_set_device()

void cuda_set_device ( int n )

This is part of the original C API. This function does nothing when Darknet was built to run on the CPU.

Here is the caller graph for this function:

◆ darknet_add_skipped_class()

void darknet_add_skipped_class	(	DarknetNetworkPtr	ptr,
		const int	class_to_skip
	)

This is the C equivalent to Darknet::add_skipped_class().

Here is the call graph for this function:

◆ darknet_clear_skipped_classes()

void darknet_clear_skipped_classes ( DarknetNetworkPtr ptr )

This is the C equivalent to Darknet::clear_skipped_classes().

Here is the call graph for this function:

◆ darknet_del_skipped_class()

void darknet_del_skipped_class	(	DarknetNetworkPtr	ptr,
		const int	class_to_include
	)

This is the C equivalent to Darknet::del_skipped_classes().

Here is the call graph for this function:

◆ darknet_fix_out_of_bound_values()

void darknet_fix_out_of_bound_values	(	DarknetNetworkPtr	ptr,
		const bool	toggle
	)

This is the C equivalent to Darknet::fix_out_of_bound_values().

Here is the call graph for this function:

◆ darknet_free_neural_network()

void darknet_free_neural_network ( DarknetNetworkPtr * ptr )

This is the C equivalent to Darknet::free_neural_network().

Here is the call graph for this function:

◆ darknet_load_neural_network()

DarknetNetworkPtr darknet_load_neural_network	(	const char *const	cfg_filename,
		const char *const	names_filename,
		const char *const	weights_filename
	)

This is the C equivalent to Darknet::load_neural_network().

Here is the call graph for this function:

◆ darknet_network_dimensions()

void darknet_network_dimensions	(	DarknetNetworkPtr	ptr,
		int *	w,
		int *	h,
		int *	c
	)

This is the C equivalent to Darknet::network_dimensions().

Here is the call graph for this function:

◆ darknet_set_detection_threshold()

void darknet_set_detection_threshold	(	DarknetNetworkPtr	ptr,
		float	threshold
	)

This is the C equivalent to Darknet::set_detection_threshold().

Here is the call graph for this function:

◆ darknet_set_gpu_index()

void darknet_set_gpu_index ( int idx )

This is the C equivalent to Darknet::set_gpu_index().

Here is the call graph for this function:

◆ darknet_set_non_maximal_suppression_threshold()

void darknet_set_non_maximal_suppression_threshold	(	DarknetNetworkPtr	ptr,
		float	threshold
	)

This is the C equivalent to Darknet::set_non_maximal_suppression_threshold().

Here is the call graph for this function:

◆ darknet_set_trace()

void darknet_set_trace ( const bool flag )

This is the C equivalent to Darknet::set_trace().

Here is the call graph for this function:

◆ darknet_set_verbose()

void darknet_set_verbose ( const bool flag )

This is the C equivalent to Darknet::set_verbose().

Here is the call graph for this function:

◆ darknet_show_version_info()

void darknet_show_version_info ( )

This is the C equivalent to Darknet::show_version_info().

Here is the call graph for this function:

◆ darknet_version_short()

const char * darknet_version_short ( )

Return the short version string, such as "3.0.163".

See also: DARKNET_VERSION_SHORT

◆ darknet_version_string()

const char * darknet_version_string ( )

Return the full version string, such as "v3.0-163-g56145bf".

See also: DARKNET_VERSION_STRING

◆ diounms_sort()

void diounms_sort	(	detection *	dets,
		int	total,
		int	classes,
		float	thresh,
		NMS_KIND	nms_kind,
		float	beta1
	)

This is part of the original C API. Non Maxima Suppression.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ do_nms_obj()

void do_nms_obj	(	detection *	dets,
		int	total,
		int	classes,
		float	thresh
	)

This is part of the original C API. Non Maxima Suppression.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ do_nms_sort()

void do_nms_sort	(	detection *	dets,
		int	total,
		int	classes,
		float	thresh
	)

This is part of the original C API. Non Maxima Suppression.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ free_detections()

void free_detections	(	detection *	dets,
		int	n
	)

This is part of the original C API. Do not use in new code.

Here is the caller graph for this function:

◆ free_image()

void free_image ( image im )

This is part of the original C API.

Free the data pointer that stores the image . All image objects must eventually call either this function or Darknet::free_image() to prevent memory leaks.

See also: Darknet::free_image()

Note: The image is passed by value, meaning that the data pointer in the caller's copy of the image will be left dangling. Be careful not to reference it once this call returns. Where possible when using C++, call Darknet::free_image() instead.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ free_network_ptr()

void free_network_ptr ( DarknetNetworkPtr ptr )

This is part of the original C API.

Do not use in new code.

See also: darknet_free_neural_network(); Darknet::free_neural_network()

If you were previously using free_network() from within C code, please use free_network_ptr() instead by passing in the address of the network structure (pointer to the network).

Note: See the additional comments in free_network().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ free_ptrs()

void free_ptrs	(	void **	ptrs,
		int	n
	)

This is part of the original C API.

Here is the caller graph for this function:

◆ get_network_boxes()

detection * get_network_boxes	(	DarknetNetworkPtr	ptr,
		int	w,
		int	h,
		float	thresh,
		float	hier,
		int *	map,
		int	relative,
		int *	num,
		int	letter
	)

This is part of the original C API.

Do not use in new code.

You must call free_detections() to free up memory once done with the detections.

See also: Darknet::predict(); free_detections()

Here is the call graph for this function:

Here is the caller graph for this function:

◆ load_image_v2()

DarknetImage load_image_v2	(	const char *	filename,
		int	desired_width,
		int	desired_height,
		int	channels
	)

This is part of the original C API. This function exists for the Python API.

Here is the call graph for this function:

◆ load_network()

DarknetNetworkPtr load_network	(	const char *	cfg,
		const char *	weights,
		int	clear
	)

This is part of the original C API. Do not use in new code.

See also: darknet_load_neural_network(); Darknet::load_neural_network()

Todo:: V3 why do we not call fuse_conv_batchnorm() here?

Here is the call graph for this function:

◆ load_network_custom()

DarknetNetworkPtr load_network_custom	(	const char *	cfg,
		const char *	weights,
		int	clear,
		int	batch
	)

Todo:: V3 Some code seems to also call this next function, and some not. This was not originally called here, but I copied it from several other code locations. Need to invetigate whether or not it should be here. 2024-08-03

Here is the call graph for this function:

Here is the caller graph for this function:

◆ make_empty_image()

DarknetImage make_empty_image	(	int	w,
		int	h,
		int	c
	)

This is part of the original C API.

Make an empty image with the given dimensions. The data pointer will be nullptr.

See also: free_image(); make_image()

Here is the caller graph for this function:

◆ make_image()

DarknetImage make_image	(	int	w,
		int	h,
		int	c
	)

This is part of the original C API.

Similar to make_empty_image() but the data pointer is fully allocated.

See also: free_image()

Here is the call graph for this function:

Here is the caller graph for this function:

◆ make_network_boxes()

detection * make_network_boxes	(	DarknetNetworkPtr	ptr,
		float	thresh,
		int *	num
	)

This is part of the original C API.

See also: make_network_boxes_batch()

Todo:: V3 JAZZ: this is where we spend all our time

Here is the call graph for this function:

Here is the caller graph for this function:

◆ network_predict_image()

float * network_predict_image	(	DarknetNetworkPtr	ptr,
		const DarknetImage	im
	)

This is part of the original C API.

Do not use in new code.

See also: network_predict_ptr(); Darknet::predict()

Here is the call graph for this function:

◆ network_predict_image_letterbox()

float * network_predict_image_letterbox	(	DarknetNetworkPtr	ptr,
		DarknetImage	im
	)

This is part of the original C API.

Here is the call graph for this function:

◆ network_predict_ptr()

float * network_predict_ptr	(	DarknetNetworkPtr	ptr,
		float *	input
	)

This is part of the original C API.

Do not use in new code.

See also: network_predict_image(); Darknet::predict()

If you were previously using network_predict() from within C code, please use network_predict_ptr() instead by passing in the address of the network structure (pointer to the network).

Here is the call graph for this function:

◆ reset_rnn()

void reset_rnn ( DarknetNetworkPtr ptr )

This is part of the original C API.

Here is the call graph for this function:

Classes

Typedefs

Enumerations

Functions

Detailed Description

Typedef Documentation

◆ box

◆ DarknetBox

◆ DarknetDetection

◆ DarknetImage

◆ DarknetNetworkPtr

◆ detection

◆ image

Enumeration Type Documentation

◆ NMS_KIND

Function Documentation

◆ calculate_binary_weights()

◆ copy_image_from_bytes()

◆ cuda_set_device()

◆ darknet_add_skipped_class()

◆ darknet_clear_skipped_classes()

◆ darknet_del_skipped_class()

◆ darknet_fix_out_of_bound_values()

◆ darknet_free_neural_network()

◆ darknet_load_neural_network()

◆ darknet_network_dimensions()

◆ darknet_set_detection_threshold()

◆ darknet_set_gpu_index()

◆ darknet_set_non_maximal_suppression_threshold()

◆ darknet_set_trace()

◆ darknet_set_verbose()

◆ darknet_show_version_info()

◆ darknet_version_short()

◆ darknet_version_string()

◆ diounms_sort()

◆ do_nms_obj()

◆ do_nms_sort()

◆ free_detections()

◆ free_image()

◆ free_network_ptr()

◆ free_ptrs()

◆ get_network_boxes()

◆ load_image_v2()

◆ load_network()

◆ load_network_custom()

◆ make_empty_image()

◆ make_image()

◆ make_network_boxes()

◆ network_predict_image()

◆ network_predict_image_letterbox()

◆ network_predict_ptr()

◆ reset_rnn()