Darknet Namespace Reference

The namespace for the C++ Darknet API. More...

Classes
class	ArgsAndParms
class	CfgAndState
class	CfgFile
	A class that represents a Darknet/YOLO configuration file. More...
class	CfgLine
	Each non-empty line in a .cfg file is stored as a Darknet::CfgLine object. More...
class	CfgSection
	A class that represents a specific section in a configuration file. More...
class	Keypoints
	The Keypoints class works with Darknet's V3 API. More...
struct	Layer
struct	Network
	Neural network structure. Contains all of the layers. Created by Darknet::CfgFile::create_network(). More...
struct	NetworkDetails
	A place to store other details related to the neural network which we cannot easily add to the usual Darknet::Network structure. More...
struct	NetworkState
struct	Output_Object
	This is used to help keep some state between calls to functions fill_network_boxes(), get_yolo_detections(), etc. More...
struct	Parm
	Structure returned by Darknet::parse_arguments(). More...
struct	Prediction
	A much-simplified version of the old API structure DarknetDetection. More...
class	TimingAndTracking
	The timing and tracking functionality is used to find places in the code where optimizations should be made. More...
class	TimingRecords
	An object of this type is statically instantiated in Timing.cpp. More...
struct	Tree

Typedefs
using	CfgLines = std::map< std::string, CfgLine >
	Lines are not stored in the order in which they are read.
using	CfgSections = std::vector< CfgSection >
using	MArgsAndParms = std::map< std::string, ArgsAndParms >
	The key is the argument name, the value is the details for that argument.
using	Output_Object_Cache = std::list< Output_Object >
using	Parms = std::vector< Parm >
	Structure returned by Darknet::parse_arguments().
using	Predictions = std::vector< Prediction >
	Each image or video frame may contain many predictions.
using	Probabilities = std::map< int, float >
	The likelyhood of a specific object class having been predicted.
using	SArgsAndParms = std::set< ArgsAndParms >
using	Skeleton = VInt
using	Skeletons = std::vector< Skeleton >
using	VThreads = std::vector< std::thread >
using	MStr = std::map< std::string, std::string >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	MStrInt = std::map< std::string, int >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	SStr = std::set< std::string >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	SInt = std::set< int >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	VFloat = std::vector< float >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	VInt = std::vector< int >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	VStr = std::vector< std::string >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	VScalars = std::vector< cv::Scalar >
	Convenient simple types used in the Darknet/YOLO C++ API.
using	NetworkPtr = DarknetNetworkPtr
	Convenient simple types used in the Darknet/YOLO C++ API.
using	Box = DarknetBox
	Convenient simple types used in the Darknet/YOLO C++ API.
using	Detection = DarknetDetection
	Convenient simple types used in the Darknet/YOLO C++ API.
using	Image = DarknetImage
	Convenient simple types used in the Darknet/YOLO C++ API.

Enumerations
enum class	EActivation {   LOGISTIC = ACTIVATION::LOGISTIC ,   RELU = ACTIVATION::RELU ,   RELU6 = ACTIVATION::RELU6 ,   RELIE = ACTIVATION::RELIE ,   LINEAR = ACTIVATION::LINEAR ,   RAMP = ACTIVATION::RAMP ,   TANH = ACTIVATION::TANH ,   PLSE = ACTIVATION::PLSE ,   REVLEAKY = ACTIVATION::REVLEAKY ,   LEAKY = ACTIVATION::LEAKY ,   ELU = ACTIVATION::ELU ,   LOGGY = ACTIVATION::LOGGY ,   STAIR = ACTIVATION::STAIR ,   HARDTAN = ACTIVATION::HARDTAN ,   LHTAN = ACTIVATION::LHTAN ,   SELU = ACTIVATION::SELU ,   GELU = ACTIVATION::GELU ,   SWISH = ACTIVATION::SWISH ,   MISH = ACTIVATION::MISH ,   HARD_MISH = ACTIVATION::HARD_MISH ,   NORM_CHAN = ACTIVATION::NORM_CHAN ,   NORM_CHAN_SOFTMAX = ACTIVATION::NORM_CHAN_SOFTMAX ,   NORM_CHAN_SOFTMAX_MAXVAL = ACTIVATION::NORM_CHAN_SOFTMAX_MAXVAL }
	The plan is to eventually remove ACTIVATION completely once we fully switch over to C++. More...
enum	EColour {   kNormal = 0 ,   kBlack ,   kRed ,   kGreen ,   kBrown ,   kBlue ,   kMagenta ,   kCyan ,   kLightGrey ,   kDarkGrey ,   kBrightRed ,   kBrightGreen ,   kYellow ,   kBrightBlue ,   kBrightMagenta ,   kBrightCyan ,   kBrightWhite }
enum class	ECostType {   SSE = COST_TYPE::SSE ,   MASKED = COST_TYPE::MASKED ,   SMOOTH = COST_TYPE::SMOOTH }
	The plan is to eventually remove COST_TYPE completely once we fully switch over to C++. More...
enum class	EIoULoss {   IOU = IOU_LOSS::IOU ,   GIOU = IOU_LOSS::GIOU ,   MSE = IOU_LOSS::MSE ,   DIOU = IOU_LOSS::DIOU ,   CIOU = IOU_LOSS::CIOU }
	The plan is to eventually remove IOU_LOSS completely once we fully switch over to C++. More...
enum class	ELayerType {   CONVOLUTIONAL ,   CONNECTED ,   MAXPOOL ,   LOCAL_AVGPOOL ,   SOFTMAX ,   DROPOUT ,   ROUTE ,   COST ,   AVGPOOL ,   SHORTCUT ,   SCALE_CHANNELS ,   SAM ,   RNN ,   LSTM ,   CRNN ,   NETWORK ,   REGION ,   YOLO ,   GAUSSIAN_YOLO ,   REORG ,   UPSAMPLE ,   EMPTY ,   BLANK ,   CONTRASTIVE ,   LAYER_LAST_IDX = ELayerType::CONTRASTIVE }
	This is the new C++ version of what used to be called LAYER_TYPE in the old C code. More...
enum class	ELearningRatePolicy {   CONSTANT = learning_rate_policy::CONSTANT ,   STEP = learning_rate_policy::STEP ,   EXP = learning_rate_policy::EXP ,   POLY = learning_rate_policy::POLY ,   STEPS = learning_rate_policy::STEPS ,   SIG = learning_rate_policy::SIG ,   RANDOM = learning_rate_policy::RANDOM ,   SGDR = learning_rate_policy::SGDR }
	The plan is to eventually remove learning_rate_policy completely once we fully switch over to C++. More...
enum class	ENMSKind {   DEFAULT_NMS = NMS_KIND::DEFAULT_NMS ,   GREEDY_NMS = NMS_KIND::GREEDY_NMS ,   DIOU_NMS = NMS_KIND::DIOU_NMS ,   CORNERS_NMS = NMS_KIND::CORNERS_NMS }
	The plan is to eventually remove NMS_KIND completely once we fully switch over to C++. More...
enum class	EParmType {   kUnknown ,   kCfgFilename ,   kNamesFilename ,   kWeightsFilename ,   kDirectory ,   kFilename ,   kOther }
	When parsing command-line parameters, each parameter is assigned a "type". More...
enum class	EWeightsNormalization {   NO_NORMALIZATION = WEIGHTS_NORMALIZATION_T::NO_NORMALIZATION ,   RELU_NORMALIZATION = WEIGHTS_NORMALIZATION_T::RELU_NORMALIZATION ,   SOFTMAX_NORMALIZATION = WEIGHTS_NORMALIZATION_T::SOFTMAX_NORMALIZATION }
	The plan is to eventually remove WEIGHTS_NORMALIZATION_T completely once we fully switch over to C++. More...
enum class	EWeightsType {   NO_WEIGHTS = WEIGHTS_TYPE_T::NO_WEIGHTS ,   PER_FEATURE = WEIGHTS_TYPE_T::PER_FEATURE ,   PER_CHANNEL = WEIGHTS_TYPE_T::PER_CHANNEL }
	The plan is to eventually remove WEIGHTS_TYPE_T completely once we fully switch over to C++. More...
enum class	EYoloPoint {   YOLO_CENTER = YOLO_POINT::YOLO_CENTER ,   YOLO_LEFT_TOP = YOLO_POINT::YOLO_LEFT_TOP ,   YOLO_RIGHT_BOTTOM = YOLO_POINT::YOLO_RIGHT_BOTTOM }
	The plan is to eventually remove YOLO_POINT completely once we fully switch over to C++. More...

Functions
SInt	add_skipped_class (Darknet::NetworkPtr ptr, const int class_to_skip)
	Add the given class index to the set of classes that Darknet must ignore.
cv::Mat	annotate (const Darknet::NetworkPtr ptr, const Predictions &predictions, cv::Mat mat)
	Annotate the given image using the predictions from Darknet::predict().
void	assign_default_class_colours (Darknet::Network *net)
	Generate the necessary class colours used to draw bounding boxes and labels.
int	best_3d_shift (const Darknet::Image &a, const Darknet::Image &b, int min, int max)
int	best_3d_shift_r (const Darknet::Image &a, const Darknet::Image &b, int min, int max)
Darknet::Image	bgr_mat_to_rgb_image (const cv::Mat &mat)
	Similar to the original mat_to_image(), but with 2 differences:
float	bilinear_interpolate (const Darknet::Image &im, float x, float y, int c)
Darknet::Image	blend_image (const Darknet::Image &fore, const Darknet::Image &back, float alpha)
void	cfg_layers ()
SInt	clear_skipped_classes (Darknet::NetworkPtr ptr)
	Clear the set of classes which Darknet must ignore.
Darknet::Image	collapse_image_layers (const Darknet::Image &source, int border)
Darknet::Image	collapse_images_horz (const Darknet::Image *ims, int n)
Darknet::Image	collapse_images_vert (const Darknet::Image *ims, int n)
void	composite_3d (char *f1, char *f2, const char *out, int delta)
void	composite_image (const Darknet::Image &source, Darknet::Image &dest, int dx, int dy)
void	constrain_image (Darknet::Image &im)
std::string	convert_to_lowercase_alphanum (const std::string &arg)
	Convert to lowercase and remove all but alphanumerics.
Darknet::Image	copy_image (const Darknet::Image &p)
void	copy_image_inplace (const Darknet::Image &src, Darknet::Image &dst)
Darknet::Image	crop_image (const Darknet::Image &im, const int dx, const int dy, const int w, const int h)
	Crop the given image.
SInt	del_skipped_class (Darknet::NetworkPtr ptr, const int class_to_include)
	Remove the given class index from the set of classes that Darknet must ignore.
char *	detection_to_json (Darknet::Detection *dets, int nboxes, int classes, const Darknet::VStr &names, long long int frame_id, char *filename)
void	display_error_msg (const std::string &msg)
	Display the given message in bright red (if colour is enabled). The message is not linefeed terminated.
void	display_usage ()
void	display_warning_msg (const std::string &msg)
	Display the given message in yellow (if colour is enabled). The message is not linefeed terminated.
void	distort_image (Darknet::Image &im, float hue, float sat, float val)
void	draw_bbox (Darknet::Image &a, const Darknet::Box &bbox, int w, float r, float g, float b)
	Draw the given bounding box. The line thickness can be specified using w.
void	draw_box (Darknet::Image &a, int x1, int y1, int x2, int y2, float r, float g, float b)
	Draw a bounding box at the rectangle coordinates within an image, using the specified colour.
void	draw_box_bw (Darknet::Image &a, int x1, int y1, int x2, int y2, float brightness)
	Draw a bounding box on a black-and-white image.
void	draw_box_width (Darknet::Image &a, int x1, int y1, int x2, int y2, int w, float r, float g, float b)
	Similiar to Darknet::draw_box(), but the line thickness can be specified using w.
void	draw_box_width_bw (Darknet::Image &a, int x1, int y1, int x2, int y2, int w, float brightness)
	Similar to Darknet::draw_box_bw(), but the line thickness can be specified using w.
void	draw_detections_v3 (Darknet::Image &im, const Darknet::Detection *dets, const int num, const float thresh, const Darknet::VStr &names, const int classes, const int ext_output)
	This is the function that is called from older Darknet code to draw annotations on an image.
void	draw_weighted_label (Darknet::Image &a, int r, int c, Darknet::Image &label, const float *rgb, const float alpha)
	Blend the label (actually an image) into the given image.
void	embed_image (const Darknet::Image &source, Darknet::Image &dest, int dx, int dy)
void	fill_image (Darknet::Image &m, float s)
void	fix_out_of_bound_values (Darknet::NetworkPtr ptr, const bool toggle)
	Fix out-of-bound values returned by Darknet::predict() for objects near the edges of images.
void	flip_image (Darknet::Image &a)
	Flip image left <-> right.
Darknet::Image	float_to_image (int w, int h, int c, float *data)
Darknet::Image	float_to_image_scaled (int w, int h, int c, float *data)
std::string	format_in_colour (const float &f, const EColour &colour, const size_t &len)
std::string	format_in_colour (const float &f, const size_t &len, const bool inverted=false)
std::string	format_in_colour (const int &i, const EColour &colour, const size_t &len)
std::string	format_in_colour (const size_t &st, const EColour &colour, const size_t &len)
std::string	format_in_colour (const std::string &str, const EColour &colour, const int &len)
std::string	format_layer_summary (const size_t idx, const Darknet::CfgSection &section, const Darknet::Layer &l)
std::string	format_loss (const double &loss)
	Format the loss combined with ANSI colours.
std::string	format_map_accuracy (const float &accuracy)
	Format the mAP% accuracy with ANSI colours.
std::string	format_map_confusion_matrix_values (const int class_id, std::string name, const float &average_precision, const int &tp, const int &fn, const int &fp, const int &tn, const float &accuracy, const float &error_rate, const float &precision, const float &recall, const float &specificity, const float &false_pos_rate)
std::string	format_time (const double &seconds_remaining)
	The time used by Darknet is a double, generated from what_time_is_it_now().
std::string	format_time_remaining (const float &seconds_remaining)
	Format the time remaining using simple-to-read text. The time must be in seconds.
void	free_data (data &d)
	Frees the "data buffer" used to load images.
void	free_image (Darknet::Image &im)
	Free the image.
void	free_neural_network (Darknet::NetworkPtr &ptr)
	Free the neural network pointer allocated in Darknet::load_neural_network().
const SArgsAndParms &	get_all_possible_arguments ()
	Get all the possible arguments used by Darknet.
const Darknet::VScalars &	get_class_colours (const Darknet::NetworkPtr ptr)
	Get access to the vector of colours assigned to each class when the .names file was loaded.
const Darknet::VStr &	get_class_names (const Darknet::NetworkPtr ptr)
	Get access to the vector of names read from the .names file when the configuration was loaded.
float	get_color (int c, int x, int max)
	Generate some "random" colour value to use. Mostly used for labels and charts.
std::string	get_command_output (const std::string &cmd)
std::filesystem::path	get_config_filename (const Darknet::NetworkPtr ptr)
	Get the filename of the configuration file that was used to load this neural network.
float	get_hierarchy_probability (float *x, Tree *hier, int c)
Darknet::Image	get_image_layer (const Darknet::Image &m, int l)
	Return a specific channel (eg: R, G, B) from an image.
std::filesystem::path	get_names_filename (const Darknet::NetworkPtr ptr)
	Get the filename of the names file that was used to load this neural network.
std::filesystem::path	get_weights_filename (const Darknet::NetworkPtr ptr)
	Get the filename of the weights file that was used to load this neural network.
Darknet::Image	grayscale_image (const Darknet::Image &im)
void	hierarchy_predictions (float *predictions, int n, Tree *hier, int only_leaves)
int	hierarchy_top_prediction (float *predictions, Tree *hier, float thresh, int stride)
void	hsv_to_rgb (Darknet::Image &im)
std::string	image_as_debug_string (const Darknet::Image &m)
	Convert the image to a debug string to display the data pointer values.
Darknet::Image	image_distance (Darknet::Image &a, Darknet::Image &b)
	Create a single-channel image for ...?
void	image_loading_loop (const int idx, load_args args)
	Run the permanent thread image loading loop.
cv::Mat	image_to_mat (const Darknet::Image &img)
	Convert the usual Darknet::Image format to OpenCV cv::Mat.
std::string	in_colour (const EColour colour)
std::string	in_colour (const EColour colour, const double d)
std::string	in_colour (const EColour colour, const float f)
std::string	in_colour (const EColour colour, const int i)
std::string	in_colour (const EColour colour, const std::string &msg)
void	initialize_new_charts (const Darknet::Network &net)
	Setup the new C++ charts. This is called once just prior to starting training.
float	iou (const cv::Rect &lhs, const cv::Rect &rhs)
	Calculate intersection-over-union given 2 OpenCV rectangles.
Darknet::Image	letterbox_image (const Darknet::Image &im, int w, int h)
void	letterbox_image_into (const Darknet::Image &im, int w, int h, Darknet::Image &boxed)
Darknet::Image	load_image (const char *filename, int desired_width=0, int desired_height=0, int channels=0)
	Load the given image.
void	load_names (Darknet::NetworkPtr net, const std::filesystem::path &filename)
	Load all of the names from a text .names file.
Darknet::NetworkPtr	load_neural_network (const std::filesystem::path &cfg_filename, const std::filesystem::path &names_filename, const std::filesystem::path &weights_filename)
	Load a neural network (.cfg) and the corresponding weights file.
Darknet::NetworkPtr	load_neural_network (Darknet::Parms &parms)
	Load a neural network.
void	load_single_image_data (load_args args)
	Load the given image data as described by the load_args parameter.
Darknet::Image	make_attention_image (int img_size, float *original_delta_cpu, float *original_input_cpu, int w, int h, int c, float alpha)
	The resulting image takes ownership of original_delta_cpu.
void	make_image_red (Darknet::Image &im)
Darknet::Image	make_random_image (int w, int h, int c)
Darknet::Image	mat_to_image (const cv::Mat &mat)
	Convert an OpenCV cv::Mat object to Darknet::Image.
void	network_dimensions (Darknet::NetworkPtr &ptr, int &w, int &h, int &c)
	Get the network dimensions (width, height, channels).
void	normalize_image (Darknet::Image &p)
void	normalize_image2 (Darknet::Image &p)
std::ostream &	operator<< (std::ostream &os, const Darknet::Prediction &pred)
	Display some information about this specific prediction.
std::ostream &	operator<< (std::ostream &os, const Darknet::Predictions &preds)
	Display some information about all the predictions.
Darknet::Parms	parse_arguments (const Darknet::VStr &v)
	Similar to the other Darknet::parse_arguments(), but uses a vector of strings as input.
Darknet::Parms	parse_arguments (int argc, char *argv[])
	Parse common Darknet command-line parameters with the values from argc and argv in main().
Predictions	predict (const Darknet::NetworkPtr ptr, const cv::Mat &mat)
	Get Darknet to look at the given image or video frame and return all predictions.
Predictions	predict (const Darknet::NetworkPtr ptr, const std::filesystem::path &image_filename)
	Get Darknet to look at the given image and return all predictions.
Predictions	predict (const Darknet::NetworkPtr ptr, Darknet::Image &img, cv::Size original_image_size=cv::Size(0, 0))
	Get Darknet to look at the given image or video frame and return all predictions.
Predictions	predict_and_annotate (const Darknet::NetworkPtr ptr, cv::Mat mat)
	Combination of Darknet::predict() and Darknet::annotate().
void	quantize_image (Darknet::Image &im)
Darknet::Image	random_augment_image (const Darknet::Image &im, const float angle, float aspect, const int low, const int high, const int size)
	Apply a bunch of random augmentations.
Darknet::Image	random_crop_image (const Darknet::Image &im, const int w, const int h)
	Similar to Darknet::crop_image() but the dx and dy are random values.
void	random_distort_image (Darknet::Image &im, float hue, float saturation, float exposure)
Tree *	read_tree (const char *filename)
Darknet::Image	resize_image (const Darknet::Image &im, int w, int h)
cv::Mat	resize_keeping_aspect_ratio (cv::Mat &mat, cv::Size desired_size, const cv::InterpolationFlags method=cv::InterpolationFlags::INTER_NEAREST)
	Resize the image as close as we can to the given size, but keep the aspect ratio the same as the original image.
Darknet::Image	resize_max (const Darknet::Image &im, int max)
Darknet::Image	resize_min (const Darknet::Image &im, int min)
cv::Mat	rgb_image_to_bgr_mat (const Darknet::Image &img)
	Similar to the original image_to_mat(), but with 2 differences:
void	rgb_to_hsv (Darknet::Image &im)
void	rgbgr_image (Darknet::Image &im)
	Do the equivalent of OpenCV's cv::COLOR_BGR2RGB to swap red and blue floats.
Darknet::Image	rotate_crop_image (const Darknet::Image &im, float rad, float s, int w, int h, float dx, float dy, float aspect)
Darknet::Image	rotate_image (const Darknet::Image &im, float rad)
	Rotate image.
void	rotate_image_cw (Darknet::Image &im, int times)
	Rotate image clockwise?
void	run_image_loading_control_thread (load_args args)
	This runs as a std::thread.
void	save_image (const Darknet::Image &p, const char *name)
void	save_image_jpg (const Darknet::Image &im, const char *name)
void	save_image_png (const Darknet::Image &im, const char *name)
void	scale_image (Darknet::Image &m, const float s)
	Scale the RGB values in an image by the given amount.
void	scale_image_channel (Darknet::Image &im, int c, float v)
void	set_annotation_draw_bb (Darknet::NetworkPtr ptr, const bool toggle)
	Determines if bounding boxes are drawn when calling either Darknet::annotate() or Darknet::predict_and_annotate().
void	set_annotation_draw_label (Darknet::NetworkPtr ptr, const bool toggle)
	Determines if text labels are drawn above the bounding boxes when calling either Darknet::annotate() or Darknet::predict_and_annotate().
void	set_annotation_font (Darknet::NetworkPtr ptr, const cv::LineTypes line_type, const cv::HersheyFonts font_face, const int font_thickness, const double font_scale)
	Set the font characteristics to use when drawing the bounding boxes and labels in either Darknet::annotate() or Darknet::predict_and_annotate().
void	set_annotation_line_type (Darknet::NetworkPtr ptr, const cv::LineTypes line_type)
	The OpenCV line type can impact performance.
const Darknet::VScalars &	set_class_colours (Darknet::NetworkPtr ptr, const Darknet::VScalars &colours)
	Set the colours to use when drawing annotations.
void	set_default_keypoints_files (Darknet::Parms &parms)
	Add the default Darknet Keypoints neural network files to the given parms.
void	set_detection_threshold (Darknet::NetworkPtr ptr, float threshold)
	Detection threshold to use when Darknet::predict() is called.
void	set_gpu_index (int idx)
	Set the GPU index to use.
void	set_non_maximal_suppression_threshold (Darknet::NetworkPtr ptr, float threshold)
	Non-maximal suppression threshold to use when Darknet::predict() is called.
void	set_rounded_corner_bounding_boxes (Darknet::NetworkPtr ptr, const bool toggle, const float roundness)
	This determines if annotations are drawn as circles or rounded rectangles in either Darknet::annotate() or Darknet::predict_and_annotate().
void	set_trace (const bool flag)
	Set the Darknet::CfgAndState::is_trace flag.
void	set_verbose (const bool flag)
	Set the Darknet::CfgAndState::is_verbose flag.
void	show_image (const Darknet::Image &p, const char *name)
void	show_image_collapsed (const Darknet::Image &p, const char *name)
void	show_image_layers (const Darknet::Image &p, const char *name)
void	show_images (Darknet::Image *ims, int n, const char *window)
void	show_version_info ()
	Display a few lines of text with some version information.
SInt	skipped_classes (const Darknet::NetworkPtr ptr)
	Return the set of classes which Darknet must ignore.
SInt	skipped_classes (Darknet::NetworkPtr ptr, const SInt &classes_to_skip)
	Set the classes which Darknet must ignore, completely over-writing all previous values.
void	stop_image_loading_threads ()
	Stop and join the image loading threads started in Darknet::run_image_loading_control_thread().
void	test_resize (char *filename)
std::string	text_to_simple_label (std::string txt)
	Convert the given text to plain alphanumeric ASCII string.
Darknet::Image	threshold_image (const Darknet::Image &im, float thresh)
void	translate_image (Darknet::Image m, float s)
void	update_accuracy_in_new_charts (const int class_index, const float accuracy)
void	update_loss_in_new_charts (const int current_iteration, const float loss, const float seconds_remaining, const bool dont_show)
	Update the new C++ charts with the given loss and mAP% accuracy value. This is called at every iteration.
std::string	trim (const std::string &str)
	Trim leading and trailing whitespace from the given string.
std::string &	trim (std::string &str)
	Trim leading and trailing whitespace from the given string.
std::string	lowercase (const std::string &str)
	Simple conversion of the string to lowercase.
std::string &	lowercase (std::string &str)
	Simple conversion of the string to lowercase.
void	dump (Darknet::Network *net, const Darknet::CfgFile::CommonParms &parms)
	Dump some information on the given network and layers to a text file for debugging.
void	dump (Darknet::CfgFile &cfg)
	Dump some information on the given network and layers to a text file for debugging.
using	NamesAndLayers = std::map< std::string, ELayerType >
	Convert between names and Darknet/YOLO layer types.
const NamesAndLayers &	all_names_and_layers ()
	Convert between names and Darknet/YOLO layer types.
ELayerType	get_layer_type_from_name (const std::string &name)
	Convert between names and Darknet/YOLO layer types.
std::string	to_string (const ELayerType type)
	Convert between names and Darknet/YOLO layer types.
using	NamesAndActivationTypes = std::map< std::string, EActivation >
	Convert between names and activation types.
const NamesAndActivationTypes &	all_names_and_activations ()
	Convert between names and activation types.
EActivation	get_activation_from_name (const std::string &name)
	Convert between names and activation types.
std::string	to_string (const EActivation activation)
	Convert between names and activation types.
using	NamesAndLearningRatePolicies = std::map< std::string, ELearningRatePolicy >
	Convert between names and learning rate policies.
const NamesAndLearningRatePolicies &	all_names_and_learning_rate_policies ()
	Convert between names and learning rate policies.
ELearningRatePolicy	get_learning_rate_policy_from_name (const std::string &name)
	Convert between names and learning rate policies.
std::string	to_string (const ELearningRatePolicy policy)
	Convert between names and learning rate policies.
using	NamesAndIoULoss = std::map< std::string, EIoULoss >
	Convert between names and IoU loss types.
const NamesAndIoULoss &	all_names_and_IoU_loss ()
	Convert between names and IoU loss types.
EIoULoss	get_IoU_loss_from_name (const std::string &name)
	Convert between names and IoU loss types.
std::string	to_string (const EIoULoss loss)
	Convert between names and IoU loss types.
using	NamesAndNMSKind = std::map< std::string, ENMSKind >
	Convert between names and NMS kind.
const NamesAndNMSKind &	all_names_and_NMS_kind ()
	Convert between names and NMS kind.
ENMSKind	get_NMS_kind_from_name (const std::string &name)
	Convert between names and NMS kind.
std::string	to_string (const ENMSKind nms_kind)
	Convert between names and NMS kind.
using	NamesAndWeightsType = std::map< std::string, EWeightsType >
	Convert between names and weights types.
const NamesAndWeightsType &	all_names_and_weights_types ()
	Convert between names and weights types.
EWeightsType	get_weights_type_from_name (const std::string &name)
	Convert between names and weights types.
std::string	to_string (const EWeightsType type)
	Convert between names and weights types.
using	NamesAndWeightsNormalization = std::map< std::string, EWeightsNormalization >
	Convert between names and weights normalization.
const NamesAndWeightsNormalization &	all_names_and_weights_normalization ()
	Convert between names and weights normalization.
EWeightsNormalization	get_weights_normalization_from_name (const std::string &name)
	Convert between names and weights normalization.
std::string	to_string (const EWeightsNormalization normalization)
	Convert between names and weights normalization.
using	NamesAndCostTypes = std::map< std::string, ECostType >
	Convert between names and cost types.
const NamesAndCostTypes &	all_names_and_cost_types ()
	Convert between names and cost types.
ECostType	get_cost_types_from_name (const std::string &name)
	Convert between names and cost types.
std::string	to_string (const ECostType type)
	Convert between names and cost types.
using	NamesAndYoloPointTypes = std::map< std::string, EYoloPoint >
	Convert between names and YOLO point types.
const NamesAndYoloPointTypes &	all_names_and_yolo_point_types ()
	Convert between names and YOLO point types.
EYoloPoint	get_yolo_point_types_from_name (const std::string &name)
	Convert between names and YOLO point types.
std::string	to_string (const EYoloPoint type)
	Convert between names and YOLO point types.

Detailed Description

The namespace for the C++ Darknet API.

Note this namespace contains both public and private API calls. The structures, enums, classes and functions declared in darknet.hpp are part of the public API.

Typedef Documentation

Box

using Darknet::Box = typedef DarknetBox

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

CfgLines

using Darknet::CfgLines = typedef std::map<std::string, CfgLine>

Lines are not stored in the order in which they are read.

Instead, they are stored as a map which allows us to quickly look up a value based on the key. There should be no duplicate keys within a section, nor should the order matter.

CfgSections

using Darknet::CfgSections = typedef std::vector<CfgSection>

Detection

using Darknet::Detection = typedef DarknetDetection

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

Image

using Darknet::Image = typedef DarknetImage

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

MArgsAndParms

using Darknet::MArgsAndParms = typedef std::map<std::string, ArgsAndParms>

The key is the argument name, the value is the details for that argument.

MStr

using Darknet::MStr = typedef std::map<std::string, std::string>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

MStrInt

using Darknet::MStrInt = typedef std::map<std::string, int>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

NamesAndActivationTypes

using Darknet::NamesAndActivationTypes = typedef std::map<std::string, EActivation>

Convert between names and activation types.

NamesAndCostTypes

using Darknet::NamesAndCostTypes = typedef std::map<std::string, ECostType>

Convert between names and cost types.

NamesAndIoULoss

using Darknet::NamesAndIoULoss = typedef std::map<std::string, EIoULoss>

Convert between names and IoU loss types.

NamesAndLayers

using Darknet::NamesAndLayers = typedef std::map<std::string, ELayerType>

Convert between names and Darknet/YOLO layer types.

NamesAndLearningRatePolicies

using Darknet::NamesAndLearningRatePolicies = typedef std::map<std::string, ELearningRatePolicy>

Convert between names and learning rate policies.

NamesAndNMSKind

using Darknet::NamesAndNMSKind = typedef std::map<std::string, ENMSKind>

Convert between names and NMS kind.

NamesAndWeightsNormalization

using Darknet::NamesAndWeightsNormalization = typedef std::map<std::string, EWeightsNormalization>

Convert between names and weights normalization.

NamesAndWeightsType

using Darknet::NamesAndWeightsType = typedef std::map<std::string, EWeightsType>

Convert between names and weights types.

NamesAndYoloPointTypes

using Darknet::NamesAndYoloPointTypes = typedef std::map<std::string, EYoloPoint>

Convert between names and YOLO point types.

NetworkPtr

using Darknet::NetworkPtr = typedef DarknetNetworkPtr

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

Output_Object_Cache

using Darknet::Output_Object_Cache = typedef std::list<Output_Object>

Parms

using Darknet::Parms = typedef std::vector<Parm>

Structure returned by Darknet::parse_arguments().

Since

2024-07-29

Predictions

using Darknet::Predictions = typedef std::vector<Prediction>

Each image or video frame may contain many predictions.

These predictions are stored in a vector in no particular order.

See also

Darknet::predict()

Since

2024-07-24

Probabilities

using Darknet::Probabilities = typedef std::map<int, float>

The likelyhood of a specific object class having been predicted.

This map contains all of the non-zero values. The key is the zero-based class indexes, and the values are the confidences for the classes, between 0.0f and 1.0f.

For example, if "car" is class #2 and "truck" is class #3, and Darknet is 95% certain it found a car and 82% certain it found a truck, the map would then contain "{ {2, 0.95}, {3, 0.82} }".

See also

Darknet::Prediction

Since

2024-07-24

SArgsAndParms

using Darknet::SArgsAndParms = typedef std::set<ArgsAndParms>

SInt

using Darknet::SInt = typedef std::set<int>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

Skeleton

using Darknet::Skeleton = typedef VInt

Skeletons

using Darknet::Skeletons = typedef std::vector<Skeleton>

SStr

using Darknet::SStr = typedef std::set<std::string>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

VFloat

using Darknet::VFloat = typedef std::vector<float>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

VInt

using Darknet::VInt = typedef std::vector<int>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

VScalars

using Darknet::VScalars = typedef std::vector<cv::Scalar>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

VStr

using Darknet::VStr = typedef std::vector<std::string>

Convenient simple types used in the Darknet/YOLO C++ API.

Since

2024-07-14

VThreads

using Darknet::VThreads = typedef std::vector<std::thread>

Enumeration Type Documentation

EActivation

enum class Darknet::EActivation

strong

The plan is to eventually remove ACTIVATION completely once we fully switch over to C++.

See also

Darknet::all_names_and_activations()

Darknet::get_activation_from_name()

Darknet::to_string()

Enumerator
LOGISTIC
RELU
RELU6
RELIE
LINEAR
RAMP
TANH
PLSE
REVLEAKY
LEAKY
ELU
LOGGY
STAIR
HARDTAN
LHTAN
SELU
GELU
SWISH
MISH
HARD_MISH
NORM_CHAN
NORM_CHAN_SOFTMAX
NORM_CHAN_SOFTMAX_MAXVAL

EColour

enum Darknet::EColour

Enumerator
kNormal
kBlack
kRed
kGreen
kBrown
kBlue
kMagenta
kCyan
kLightGrey
kDarkGrey
kBrightRed
kBrightGreen
kYellow
kBrightBlue
kBrightMagenta
kBrightCyan
kBrightWhite

ECostType

enum class Darknet::ECostType

strong

The plan is to eventually remove COST_TYPE completely once we fully switch over to C++.

See also

Darknet::all_names_and_cost_types()

Darknet::get_cost_types_from_name()

Darknet::to_string()

Enumerator
SSE
MASKED
SMOOTH

EIoULoss

enum class Darknet::EIoULoss

strong

The plan is to eventually remove IOU_LOSS completely once we fully switch over to C++.

See also

Darknet::all_names_and_IoU_loss()

Darknet::get_IoU_loss_from_name()

Darknet::to_string()

Enumerator
IOU
GIOU
MSE
DIOU
CIOU

ELayerType

enum class Darknet::ELayerType

strong

This is the new C++ version of what used to be called LAYER_TYPE in the old C code.

See also

Darknet::all_names_and_layers()

Darknet::get_layer_type_from_name()

Darknet::to_string()

Enumerator
CONVOLUTIONAL	this is used everywhere
CONNECTED	used in 3 rarely used configurations
MAXPOOL	used often
LOCAL_AVGPOOL	only used in yolov4-tiny_contrastive.cfg – obsolete?
SOFTMAX	used in several non-YOLO configs
DROPOUT	used in 3 non-YOLO configs
ROUTE	used often
COST	used in several non-YOLO configs
AVGPOOL	used in several non-YOLO configs
SHORTCUT	used almost everywhere
SCALE_CHANNELS	only used in 2 configs
SAM	only used in 3 configs
RNN	only used in rnn.train.cfg – obsolete?
LSTM	only used in lstm.train.cfg – obsolete?
CRNN	only used in 2 non-YOLO configs
NETWORK	used in every config
REGION	only used in tiny-yolo_xnor.cfg (which is NOT a YOLO config)
YOLO	used often
GAUSSIAN_YOLO	only used in Gaussian_yolov3_BDD.cfg
REORG	only used in yolov4-sam-mish-csp-reorg-bfm.cfg
UPSAMPLE	used often, does downsampling instead if l.reverse=1
EMPTY	UNUSED
BLANK	UNUSED
CONTRASTIVE	only used in yolov4-tiny_contrastive.cfg
LAYER_LAST_IDX	point to the last-used idx

ELearningRatePolicy

enum class Darknet::ELearningRatePolicy

strong

The plan is to eventually remove learning_rate_policy completely once we fully switch over to C++.

See also

Darknet::all_names_and_learning_rate_policies()

Darknet::get_learning_rate_policy_from_name()

Darknet::to_string()

Enumerator
CONSTANT
STEP
EXP
POLY
STEPS
SIG
RANDOM
SGDR

ENMSKind

enum class Darknet::ENMSKind

strong

The plan is to eventually remove NMS_KIND completely once we fully switch over to C++.

See also

Darknet::all_names_and_NMS_kind()

Darknet::get_NMS_kind_from_name()

Darknet::to_string()

Enumerator
DEFAULT_NMS
GREEDY_NMS
DIOU_NMS
CORNERS_NMS

EParmType

enum class Darknet::EParmType

strong

When parsing command-line parameters, each parameter is assigned a "type".

See also

Darknet::Parm

Darknet::parse_arguments()

Since

2024-07-29

Enumerator
kUnknown	Should be unused. See "kOther" instead for all other parameter types.
kCfgFilename	The configuration file to load. There should only be 1 parameter of this type.
kNamesFilename	The names file to load. There should only be 1 parameter of this type.
kWeightsFilename	The weights file to load. There should only be 1 parameter of this type.
kDirectory	An existing directory name.
kFilename	An existing filename which is not one of the 3 neural network files.
kOther	Any other parameter.

EWeightsNormalization

enum class Darknet::EWeightsNormalization

strong

The plan is to eventually remove WEIGHTS_NORMALIZATION_T completely once we fully switch over to C++.

See also

Darknet::all_names_and_weights_normalization()

Darknet::get_weights_normalization_from_name()

Darknet::to_string()

Enumerator
NO_NORMALIZATION
RELU_NORMALIZATION
SOFTMAX_NORMALIZATION

EWeightsType

enum class Darknet::EWeightsType

strong

The plan is to eventually remove WEIGHTS_TYPE_T completely once we fully switch over to C++.

See also

Darknet::all_names_and_weights_types()

Darknet::get_weights_type_from_name()

Darknet::to_string()

Enumerator
NO_WEIGHTS
PER_FEATURE
PER_CHANNEL

EYoloPoint

enum class Darknet::EYoloPoint

strong

The plan is to eventually remove YOLO_POINT completely once we fully switch over to C++.

See also

Darknet::all_names_and_yolo_point_types()

Darknet::get_yolo_point_types_from_name()

Darknet::to_string()

Enumerator
YOLO_CENTER
YOLO_LEFT_TOP
YOLO_RIGHT_BOTTOM

Function Documentation

add_skipped_class()

Darknet::SInt Darknet::add_skipped_class	(	Darknet::NetworkPtr	ptr,
		const int	class_to_skip
	)

Add the given class index to the set of classes that Darknet must ignore.

This may be called multiple times if you have many classes you want skipped, or you can call skipped_classes() if you want to set them all at once.

See also

Darknet::skipped_classes()

Darknet::clear_skipped_classes()

Darknet::del_skipped_class()

Since

2024-10-07

Here is the caller graph for this function:

all_names_and_activations()

const Darknet::NamesAndActivationTypes & Darknet::all_names_and_activations

(

)

Convert between names and activation types.

Here is the caller graph for this function:

all_names_and_cost_types()

const Darknet::NamesAndCostTypes & Darknet::all_names_and_cost_types

(

)

Convert between names and cost types.

Here is the caller graph for this function:

all_names_and_IoU_loss()

const Darknet::NamesAndIoULoss & Darknet::all_names_and_IoU_loss

(

)

Convert between names and IoU loss types.

Here is the caller graph for this function:

all_names_and_layers()

const Darknet::NamesAndLayers & Darknet::all_names_and_layers

(

)

Convert between names and Darknet/YOLO layer types.

Here is the caller graph for this function:

all_names_and_learning_rate_policies()

const Darknet::NamesAndLearningRatePolicies & Darknet::all_names_and_learning_rate_policies

(

)

Convert between names and learning rate policies.

Here is the caller graph for this function:

all_names_and_NMS_kind()

const Darknet::NamesAndNMSKind & Darknet::all_names_and_NMS_kind

(

)

Convert between names and NMS kind.

Here is the caller graph for this function:

all_names_and_weights_normalization()

const Darknet::NamesAndWeightsNormalization & Darknet::all_names_and_weights_normalization

(

)

Convert between names and weights normalization.

Here is the caller graph for this function:

all_names_and_weights_types()

const Darknet::NamesAndWeightsType & Darknet::all_names_and_weights_types

(

)

Convert between names and weights types.

Here is the caller graph for this function:

all_names_and_yolo_point_types()

const Darknet::NamesAndYoloPointTypes & Darknet::all_names_and_yolo_point_types

(

)

Convert between names and YOLO point types.

Here is the caller graph for this function:

annotate()

cv::Mat Darknet::annotate	(	const Darknet::NetworkPtr	ptr,
		const Predictions &	predictions,
		cv::Mat	mat
	)

Annotate the given image using the predictions from Darknet::predict().

See also

Darknet::predict_and_annotate()

Since

2024-07-30

Here is the caller graph for this function:

assign_default_class_colours()

void Darknet::assign_default_class_colours

(

Darknet::Network *

net

)

Generate the necessary class colours used to draw bounding boxes and labels.

The colours are stored in Darknet::NetworkDetails at the time the network is loaded, once the total number of classes are known.

This is called automatically when the network is initialized via Darknet::CfgFile. There is no need to manually call this function.

Since

2024-08-06

Here is the call graph for this function:

Here is the caller graph for this function:

best_3d_shift()

int Darknet::best_3d_shift	(	const Darknet::Image &	a,
		const Darknet::Image &	b,
		int	min,
		int	max
	)

Here is the call graph for this function:

best_3d_shift_r()

int Darknet::best_3d_shift_r	(	const Darknet::Image &	a,
		const Darknet::Image &	b,
		int	min,
		int	max
	)

Here is the call graph for this function:

Here is the caller graph for this function:

bgr_mat_to_rgb_image()

Darknet::Image Darknet::bgr_mat_to_rgb_image

(

const cv::Mat &

mat

)

Similar to the original mat_to_image(), but with 2 differences:

• the input image is in the "natural" OpenCV BGR format (the output image is still in RGB format),
• this function uses very efficent OpenCV techniques to convert the cv::Mat to Darknet::Image which makes it much faster.

Remember to call Darknet::free_image() when done.

See also

Darknet::rgb_image_to_bgr_mat()

Since

2024-08-23

Todo:

COLOR this function assumes 3-channel images

Here is the call graph for this function:

Here is the caller graph for this function:

bilinear_interpolate()

float Darknet::bilinear_interpolate	(	const Darknet::Image &	im,
		float	x,
		float	y,
		int	c
	)

Here is the caller graph for this function:

blend_image()

Darknet::Image Darknet::blend_image	(	const Darknet::Image &	fore,
		const Darknet::Image &	back,
		float	alpha
	)

Here is the call graph for this function:

cfg_layers()

void Darknet::cfg_layers

(

)

Here is the call graph for this function:

Here is the caller graph for this function:

clear_skipped_classes()

Darknet::SInt Darknet::clear_skipped_classes

(

Darknet::NetworkPtr

ptr

)

Clear the set of classes which Darknet must ignore.

The default is for Darknet/YOLO to not skip any classes, as if this function has been called.

See also

Darknet::skipped_classes()

Darknet::add_skipped_class()

Darknet::del_skipped_class()

Since

2024-10-07

Here is the caller graph for this function:

collapse_image_layers()

Darknet::Image Darknet::collapse_image_layers	(	const Darknet::Image &	source,
		int	border
	)

Here is the call graph for this function:

Here is the caller graph for this function:

collapse_images_horz()

Darknet::Image Darknet::collapse_images_horz	(	const Darknet::Image *	ims,
		int	n
	)

collapse_images_vert()

Darknet::Image Darknet::collapse_images_vert	(	const Darknet::Image *	ims,
		int	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

composite_3d()

void Darknet::composite_3d	(	char *	f1,
		char *	f2,
		const char *	out,
		int	delta
	)

Here is the call graph for this function:

Here is the caller graph for this function:

composite_image()

void Darknet::composite_image	(	const Darknet::Image &	source,
		Darknet::Image &	dest,
		int	dx,
		int	dy
	)

Here is the call graph for this function:

constrain_image()

void Darknet::constrain_image

(

Darknet::Image &

im

)

Here is the caller graph for this function:

convert_to_lowercase_alphanum()

std::string Darknet::convert_to_lowercase_alphanum

(

const std::string &

arg

)

Convert to lowercase and remove all but alphanumerics.

Here is the caller graph for this function:

copy_image()

Darknet::Image Darknet::copy_image

(

const Darknet::Image &

p

)

Here is the call graph for this function:

Here is the caller graph for this function:

copy_image_inplace()

void Darknet::copy_image_inplace	(	const Darknet::Image &	src,
		Darknet::Image &	dst
	)

Note

Function currently seems to be unused.

crop_image()

Darknet::Image Darknet::crop_image	(	const Darknet::Image &	im,
		const int	dx,
		const int	dy,
		const int	w,
		const int	h
	)

Crop the given image.

See also

Darknet::random_crop_image()

Here is the call graph for this function:

Here is the caller graph for this function:

del_skipped_class()

Darknet::SInt Darknet::del_skipped_class	(	Darknet::NetworkPtr	ptr,
		const int	class_to_include
	)

Remove the given class index from the set of classes that Darknet must ignore.

This may be called multiple times if there are several class indexes you'll like to restore.

See also

Darknet::skipped_classes()

Darknet::clear_skipped_classes()

Darknet::add_skipped_class()

Since

2024-10-07

Here is the caller graph for this function:

detection_to_json()

char * Darknet::detection_to_json	(	Darknet::Detection *	dets,
		int	nboxes,
		int	classes,
		const Darknet::VStr &	names,
		long long int	frame_id,
		char *	filename
	)

Here is the caller graph for this function:

display_error_msg()

void Darknet::display_error_msg

(

const std::string &

msg

)

Display the given message in bright red (if colour is enabled). The message is not linefeed terminated.

Here is the call graph for this function:

Here is the caller graph for this function:

display_usage()

void Darknet::display_usage

(

)

Here is the call graph for this function:

Here is the caller graph for this function:

display_warning_msg()

void Darknet::display_warning_msg

(

const std::string &

msg

)

Display the given message in yellow (if colour is enabled). The message is not linefeed terminated.

Here is the call graph for this function:

Here is the caller graph for this function:

distort_image()

void Darknet::distort_image	(	Darknet::Image &	im,
		float	hue,
		float	sat,
		float	val
	)

Todo:

COLOR - needs to be fixed for 1 <= c <= N

Here is the call graph for this function:

Here is the caller graph for this function:

draw_bbox()

void Darknet::draw_bbox	(	Darknet::Image &	a,
		const Darknet::Box &	bbox,
		int	w,
		float	r,
		float	g,
		float	b
	)

Draw the given bounding box. The line thickness can be specified using w.

Here is the call graph for this function:

draw_box()

void Darknet::draw_box	(	Darknet::Image &	a,
		int	x1,
		int	y1,
		int	x2,
		int	y2,
		float	r,
		float	g,
		float	b
	)

Draw a bounding box at the rectangle coordinates within an image, using the specified colour.

Here is the caller graph for this function:

draw_box_bw()

void Darknet::draw_box_bw	(	Darknet::Image &	a,
		int	x1,
		int	y1,
		int	x2,
		int	y2,
		float	brightness
	)

Draw a bounding box on a black-and-white image.

Here is the caller graph for this function:

draw_box_width()

void Darknet::draw_box_width	(	Darknet::Image &	a,
		int	x1,
		int	y1,
		int	x2,
		int	y2,
		int	w,
		float	r,
		float	g,
		float	b
	)

Similiar to Darknet::draw_box(), but the line thickness can be specified using w.

Here is the call graph for this function:

Here is the caller graph for this function:

draw_box_width_bw()

void Darknet::draw_box_width_bw	(	Darknet::Image &	a,
		int	x1,
		int	y1,
		int	x2,
		int	y2,
		int	w,
		float	brightness
	)

Similar to Darknet::draw_box_bw(), but the line thickness can be specified using w.

Here is the call graph for this function:

Here is the caller graph for this function:

draw_detections_v3()

void Darknet::draw_detections_v3	(	Darknet::Image &	im,
		const Darknet::Detection *	dets,
		const int	num,
		const float	thresh,
		const Darknet::VStr &	names,
		const int	classes,
		const int	ext_output
	)

This is the function that is called from older Darknet code to draw annotations on an image.

Todo:

replace qsort() mid priority

Todo:

replace qsort() mid priority

Here is the call graph for this function:

Here is the caller graph for this function:

draw_weighted_label()

void Darknet::draw_weighted_label	(	Darknet::Image &	a,
		int	r,
		int	c,
		Darknet::Image &	label,
		const float *	rgb,
		const float	alpha
	)

Blend the label (actually an image) into the given image.

Here is the call graph for this function:

Here is the caller graph for this function:

dump() [1/2]

void Darknet::dump

(

Darknet::CfgFile &

cfg

)

Dump some information on the given network and layers to a text file for debugging.

Here is the call graph for this function:

dump() [2/2]

void Darknet::dump	(	Darknet::Network *	net,
		const Darknet::CfgFile::CommonParms &	parms
	)

Dump some information on the given network and layers to a text file for debugging.

Here is the call graph for this function:

Here is the caller graph for this function:

embed_image()

void Darknet::embed_image	(	const Darknet::Image &	source,
		Darknet::Image &	dest,
		int	dx,
		int	dy
	)

Here is the call graph for this function:

Here is the caller graph for this function:

fill_image()

void Darknet::fill_image	(	Darknet::Image &	m,
		float	s
	)

Here is the caller graph for this function:

fix_out_of_bound_values()

void Darknet::fix_out_of_bound_values	(	Darknet::NetworkPtr	ptr,
		const bool	toggle
	)

Fix out-of-bound values returned by Darknet::predict() for objects near the edges of images.

When set to true, this will ensure that normalized coordinates are between 0.0 and 1.0, and do not extend beyond the borders of the image or video frame.

Default is true.

See also

Darknet::NetworkDetails::fix_out_of_bound_normalized_coordinates

Since

2024-07-25

Here is the caller graph for this function:

flip_image()

void Darknet::flip_image

(

Darknet::Image &

a

)

Flip image left <-> right.

Here is the caller graph for this function:

float_to_image()

Darknet::Image Darknet::float_to_image	(	int	w,
		int	h,
		int	c,
		float *	data
	)

Here is the call graph for this function:

Here is the caller graph for this function:

float_to_image_scaled()

Darknet::Image Darknet::float_to_image_scaled	(	int	w,
		int	h,
		int	c,
		float *	data
	)

Here is the call graph for this function:

Here is the caller graph for this function:

format_in_colour() [1/5]

std::string Darknet::format_in_colour	(	const float &	f,
		const EColour &	colour,
		const size_t &	len
	)

Here is the call graph for this function:

format_in_colour() [2/5]

std::string Darknet::format_in_colour	(	const float &	f,
		const size_t &	len,
		const bool	inverted = false
	)

Here is the call graph for this function:

format_in_colour() [3/5]

std::string Darknet::format_in_colour	(	const int &	i,
		const EColour &	colour,
		const size_t &	len
	)

Here is the call graph for this function:

format_in_colour() [4/5]

std::string Darknet::format_in_colour	(	const size_t &	st,
		const EColour &	colour,
		const size_t &	len
	)

Here is the call graph for this function:

format_in_colour() [5/5]

std::string Darknet::format_in_colour	(	const std::string &	str,
		const EColour &	colour,
		const int &	len
	)

Here is the call graph for this function:

Here is the caller graph for this function:

format_layer_summary()

std::string Darknet::format_layer_summary	(	const size_t	idx,
		const Darknet::CfgSection &	section,
		const Darknet::Layer &	l
	)

Here is the call graph for this function:

Here is the caller graph for this function:

format_loss()

std::string Darknet::format_loss

(

const double &

loss

)

Format the loss combined with ANSI colours.

Here is the call graph for this function:

Here is the caller graph for this function:

format_map_accuracy()

std::string Darknet::format_map_accuracy

(

const float &

accuracy

)

Format the mAP% accuracy with ANSI colours.

Here is the call graph for this function:

Here is the caller graph for this function:

format_map_confusion_matrix_values()

std::string Darknet::format_map_confusion_matrix_values	(	const int	class_id,
		std::string	name,
		const float &	average_precision,
		const int &	tp,
		const int &	fn,
		const int &	fp,
		const int &	tn,
		const float &	accuracy,
		const float &	error_rate,
		const float &	precision,
		const float &	recall,
		const float &	specificity,
		const float &	false_pos_rate
	)

Here is the call graph for this function:

Here is the caller graph for this function:

format_time()

std::string Darknet::format_time

(

const double &

seconds_remaining

)

The time used by Darknet is a double, generated from what_time_is_it_now().

It is the number of seconds since epoch with milliseconds and microseconds as decimals. This function will format one of these double using the most intelligent unit necessary.

Here is the caller graph for this function:

format_time_remaining()

std::string Darknet::format_time_remaining

(

const float &

seconds_remaining

)

Format the time remaining using simple-to-read text. The time must be in seconds.

Here is the caller graph for this function:

free_data()

void Darknet::free_data

(

data &

d

)

Frees the "data buffer" used to load images.

Here is the call graph for this function:

Here is the caller graph for this function:

free_image()

void Darknet::free_image

(

Darknet::Image &

im

)

Free the image.

Unlike the C version free_image(), the image object is passed by reference, so the pointer to the data will be reset to nullptr once the memory has been freed. When possible, prefer calling Darknet::free_image().

All images must call either free_image() or Darknet::free_image() to avoid memory leaks.

free_neural_network()

void Darknet::free_neural_network

(

Darknet::NetworkPtr &

ptr

)

Free the neural network pointer allocated in Darknet::load_neural_network().

Does nothing if the pointer has already been freed. Will reset the pointer to nullptr once the structure has been freed.

You should call this once you are done with Darknet to avoid memory leaks in your application.

Since

2024-07-25

Here is the call graph for this function:

Here is the caller graph for this function:

get_activation_from_name()

Darknet::EActivation Darknet::get_activation_from_name

(

const std::string &

name

)

Convert between names and activation types.

Here is the call graph for this function:

Here is the caller graph for this function:

get_all_possible_arguments()

const Darknet::SArgsAndParms & Darknet::get_all_possible_arguments

(

)

Get all the possible arguments used by Darknet.

This is not what the user specified to main() but all the possible arguments against which we validate the input.

Here is the caller graph for this function:

get_class_colours()

const Darknet::VScalars & Darknet::get_class_colours

(

const Darknet::NetworkPtr

ptr

)

Get access to the vector of colours assigned to each class when the .names file was loaded.

Since

2024-08-06

get_class_names()

const Darknet::VStr & Darknet::get_class_names

(

const Darknet::NetworkPtr

ptr

)

Get access to the vector of names read from the .names file when the configuration was loaded.

Since

2024-08-06

Here is the caller graph for this function:

get_color()

float Darknet::get_color	(	int	c,
		int	x,
		int	max
	)

Generate some "random" colour value to use. Mostly used for labels and charts.

Here is the caller graph for this function:

get_command_output()

std::string Darknet::get_command_output

(

const std::string &

cmd

)

Here is the caller graph for this function:

get_config_filename()

std::filesystem::path Darknet::get_config_filename

(

const Darknet::NetworkPtr

ptr

)

Get the filename of the configuration file that was used to load this neural network.

Since

2024-08-29

Here is the caller graph for this function:

get_cost_types_from_name()

Darknet::ECostType Darknet::get_cost_types_from_name

(

const std::string &

name

)

Convert between names and cost types.

Here is the call graph for this function:

Here is the caller graph for this function:

get_hierarchy_probability()

float Darknet::get_hierarchy_probability	(	float *	x,
		Darknet::Tree *	hier,
		int	c
	)

get_image_layer()

Darknet::Image Darknet::get_image_layer	(	const Darknet::Image &	m,
		int	l
	)

Return a specific channel (eg: R, G, B) from an image.

Here is the call graph for this function:

Here is the caller graph for this function:

get_IoU_loss_from_name()

Darknet::EIoULoss Darknet::get_IoU_loss_from_name

(

const std::string &

name

)

Convert between names and IoU loss types.

Here is the call graph for this function:

Here is the caller graph for this function:

get_layer_type_from_name()

Darknet::ELayerType Darknet::get_layer_type_from_name

(

const std::string &

name

)

Convert between names and Darknet/YOLO layer types.

Here is the call graph for this function:

get_learning_rate_policy_from_name()

Darknet::ELearningRatePolicy Darknet::get_learning_rate_policy_from_name

(

const std::string &

name

)

Convert between names and learning rate policies.

Here is the call graph for this function:

Here is the caller graph for this function:

get_names_filename()

std::filesystem::path Darknet::get_names_filename

(

const Darknet::NetworkPtr

ptr

)

Get the filename of the names file that was used to load this neural network.

Since

2024-08-29

get_NMS_kind_from_name()

Darknet::ENMSKind Darknet::get_NMS_kind_from_name

(

const std::string &

name

)

Convert between names and NMS kind.

Here is the call graph for this function:

Here is the caller graph for this function:

get_weights_filename()

std::filesystem::path Darknet::get_weights_filename

(

const Darknet::NetworkPtr

ptr

)

Get the filename of the weights file that was used to load this neural network.

Since

2024-08-29

Here is the caller graph for this function:

get_weights_normalization_from_name()

Darknet::EWeightsNormalization Darknet::get_weights_normalization_from_name

(

const std::string &

name

)

Convert between names and weights normalization.

Here is the call graph for this function:

Here is the caller graph for this function:

get_weights_type_from_name()

Darknet::EWeightsType Darknet::get_weights_type_from_name

(

const std::string &

name

)

Convert between names and weights types.

Here is the call graph for this function:

Here is the caller graph for this function:

get_yolo_point_types_from_name()

Darknet::EYoloPoint Darknet::get_yolo_point_types_from_name

(

const std::string &

name

)

Convert between names and YOLO point types.

Here is the call graph for this function:

Here is the caller graph for this function:

grayscale_image()

Darknet::Image Darknet::grayscale_image

(

const Darknet::Image &

im

)

Here is the call graph for this function:

Here is the caller graph for this function:

hierarchy_predictions()

void Darknet::hierarchy_predictions	(	float *	predictions,
		int	n,
		Darknet::Tree *	hier,
		int	only_leaves
	)

hierarchy_top_prediction()

int Darknet::hierarchy_top_prediction	(	float *	predictions,
		Darknet::Tree *	hier,
		float	thresh,
		int	stride
	)

hsv_to_rgb()

void Darknet::hsv_to_rgb

(

Darknet::Image &

im

)

Todo:

COLOR - cannot do HSV if channels > 3

Here is the call graph for this function:

Here is the caller graph for this function:

image_as_debug_string()

std::string Darknet::image_as_debug_string

(

const Darknet::Image &

m

)

Convert the image to a debug string to display the data pointer values.

For example, a tiny 5x3 image might look like this:

Darknet::Image: 5x3x3, data=0x5858cb1514e0
R 0000: 1.00 1.00 1.00 1.00 1.00
R 0005: 1.00 1.00 1.00 0.13 0.13
R 0010: 1.00 1.00 1.00 0.13 0.13
G 0015: 0.25 0.25 0.25 0.25 0.25
G 0020: 0.25 0.25 0.25 0.13 0.13
G 0025: 0.25 0.25 0.25 0.13 0.13
B 0030: 0.50 0.50 0.50 0.50 0.50
B 0035: 0.50 0.50 0.50 1.00 1.00
B 0040: 0.50 0.50 0.50 1.00 1.00

image_distance()

Darknet::Image Darknet::image_distance	(	Darknet::Image &	a,
		Darknet::Image &	b
	)

Create a single-channel image for ...?

Note

Function currently seems to be unused.

Here is the call graph for this function:

image_loading_loop()

void Darknet::image_loading_loop	(	const int	idx,
		load_args	args
	)

Run the permanent thread image loading loop.

This is started by Darknet::run_image_loading_control_thread(), and is stopped by Darknet::stop_image_loading_threads().

This was originally called run_thread_loop() and used pthread, but has since been re-written to use C++11.

Since

2024-04-02

Todo:

get rid of this busy-loop

Here is the call graph for this function:

Here is the caller graph for this function:

image_to_mat()

cv::Mat Darknet::image_to_mat

(

const Darknet::Image &

img

)

Convert the usual Darknet::Image format to OpenCV cv::Mat.

The mat object will be in RGB format, not BGR.

See also

cv::COLOR_RGB2BGR

Darknet::mat_to_image()

Here is the caller graph for this function:

in_colour() [1/5]

std::string Darknet::in_colour

(

const EColour

colour

)

in_colour() [2/5]

std::string Darknet::in_colour	(	const EColour	colour,
		const double	d
	)

Here is the call graph for this function:

in_colour() [3/5]

std::string Darknet::in_colour	(	const EColour	colour,
		const float	f
	)

Here is the call graph for this function:

in_colour() [4/5]

std::string Darknet::in_colour	(	const EColour	colour,
		const int	i
	)

Here is the call graph for this function:

Here is the caller graph for this function:

in_colour() [5/5]

std::string Darknet::in_colour	(	const EColour	colour,
		const std::string &	msg
	)

initialize_new_charts()

void Darknet::initialize_new_charts

(

const Darknet::Network &

net

)

Setup the new C++ charts. This is called once just prior to starting training.

See also

Chart

Here is the caller graph for this function:

iou()

float Darknet::iou	(	const cv::Rect &	lhs,
		const cv::Rect &	rhs
	)

Calculate intersection-over-union given 2 OpenCV rectangles.

Will return a value between 0.0f and 1.0f.

See also

box_iou()

Since

2024-09-07

Here is the caller graph for this function:

letterbox_image()

Darknet::Image Darknet::letterbox_image	(	const Darknet::Image &	im,
		int	w,
		int	h
	)

Here is the call graph for this function:

Here is the caller graph for this function:

letterbox_image_into()

void Darknet::letterbox_image_into	(	const Darknet::Image &	im,
		int	w,
		int	h,
		Darknet::Image &	boxed
	)

Here is the call graph for this function:

load_image()

Darknet::Image Darknet::load_image	(	const char *	filename,
		int	desired_width = 0,
		int	desired_height = 0,
		int	channels = 0
	)

Load the given image.

If both desired_width and desired_height have been set, then the image will be resized to match those dimensions. Otherwise, specify 0 (zero) to leave the image dimensions unchanged.

Unless single channel greyscale has been requested, the image returned will be in RGB format, not BGR.

Note

Remember to call Darknet::free_image() once an image is no longer needed.

See also

Darknet::image_to_mat()

Darknet::mat_to_image()

Here is the call graph for this function:

Here is the caller graph for this function:

load_names()

void Darknet::load_names	(	Darknet::NetworkPtr	net,
		const std::filesystem::path &	filename
	)

Load all of the names from a text .names file.

The file must exist, and must have the same number of classes as the network.

This will automatically be called if you use Darknet::load_neural_network() from the new recommended V3 API.

You may need to manually call it if you use the old load_network() or load_network_custom() from the original C API. If you don't call it, some default placeholders will be used instead, such as "#0", "#1", etc.

Since

2024-08-06

Here is the call graph for this function:

Here is the caller graph for this function:

load_neural_network() [1/2]

Darknet::NetworkPtr Darknet::load_neural_network	(	const std::filesystem::path &	cfg_filename,
		const std::filesystem::path &	names_filename,
		const std::filesystem::path &	weights_filename
	)

Load a neural network (.cfg) and the corresponding weights file.

Remember to call Darknet::free_neural_network() once the neural network is no longer needed.

Since

2024-07-24

Here is the call graph for this function:

Here is the caller graph for this function:

load_neural_network() [2/2]

Darknet::NetworkPtr Darknet::load_neural_network

(

Darknet::Parms &

parms

)

Load a neural network.

Remember to call Darknet::free_neural_network() once the neural network is no longer needed.

See also

Darknet::parse_arguments()

Since

2024-07-29

Here is the call graph for this function:

load_single_image_data()

void Darknet::load_single_image_data

(

load_args

args

)

Load the given image data as described by the load_args parameter.

This is typically used to load images on a secondary thread, such as image_loading_loop().

Note

The name is misleading. While I initially thought a single image at a time was being loaded, the args.n argument is used to describe the number of images that will be loaded together. This will typically be the batch size divided by the number of worker threads (default is 6 threads).

This was originally called load_thread().

Since

2024-04-02

Here is the call graph for this function:

Here is the caller graph for this function:

lowercase() [1/2]

std::string Darknet::lowercase

(

const std::string &

str

)

Simple conversion of the string to lowercase.

Here is the call graph for this function:

Here is the caller graph for this function:

lowercase() [2/2]

std::string & Darknet::lowercase

(

std::string &

str

)

Simple conversion of the string to lowercase.

make_attention_image()

Darknet::Image Darknet::make_attention_image	(	int	img_size,
		float *	original_delta_cpu,
		float *	original_input_cpu,
		int	w,
		int	h,
		int	c,
		float	alpha
	)

The resulting image takes ownership of original_delta_cpu.

Here is the call graph for this function:

Here is the caller graph for this function:

make_image_red()

void Darknet::make_image_red

(

Darknet::Image &

im

)

Here is the call graph for this function:

Here is the caller graph for this function:

make_random_image()

Darknet::Image Darknet::make_random_image	(	int	w,
		int	h,
		int	c
	)

Here is the call graph for this function:

Here is the caller graph for this function:

mat_to_image()

Darknet::Image Darknet::mat_to_image

(

const cv::Mat &

mat

)

Convert an OpenCV cv::Mat object to Darknet::Image.

The cv::Mat is expected to already have been converted from BGR to RGB. The result Darknet::Image floats will be normalized between 0.0 and 1.0. Remember to call Darknet::free_image() when done.

See also

Darknet::bgr_mat_to_rgb_image()

Darknet::rgb_image_to_bgr_mat()

Darknet::image_to_mat()

cv::COLOR_BGR2RGB

Here is the call graph for this function:

Here is the caller graph for this function:

network_dimensions()

void Darknet::network_dimensions	(	Darknet::NetworkPtr &	ptr,
		int &	w,
		int &	h,
		int &	c
	)

Get the network dimensions (width, height, channels).

Since

2024-07-25

Here is the caller graph for this function:

normalize_image()

void Darknet::normalize_image

(

Darknet::Image &

p

)

Here is the caller graph for this function:

normalize_image2()

void Darknet::normalize_image2

(

Darknet::Image &

p

)

Note

Function currently seems to be unused.

operator<<() [1/2]

std::ostream & Darknet::operator<<	(	std::ostream &	os,
		const Darknet::Prediction &	pred
	)

Display some information about this specific prediction.

Use like this:

std::cout << prediction << std::endl;

Since

2024-08-06

Here is the caller graph for this function:

operator<<() [2/2]

std::ostream & Darknet::operator<<	(	std::ostream &	os,
		const Darknet::Predictions &	preds
	)

Display some information about all the predictions.

Use like this:

const auto results = Darknet::predict(ptr, mat);
std::cout << results << std::endl;

Output would look similar to this:

prediction results: 5
-> 1/5: #4 prob=0.999923 x=423 y=35 w=206 h=221 entries=1
-> 2/5: #3 prob=0.999828 x=285 y=85 w=138 h=135 entries=1
-> 3/5: #2 prob=0.988380 x=527 y=128 w=31 h=28 entries=1
-> 4/5: #1 prob=0.996240 x=498 y=187 w=26 h=29 entries=1
-> 5/5: #0 prob=0.994430 x=46 y=127 w=43 h=40 entries=1

Since

2024-08-06

Here is the call graph for this function:

parse_arguments() [1/2]

Darknet::Parms Darknet::parse_arguments

(

const Darknet::VStr &

v

)

Similar to the other Darknet::parse_arguments(), but uses a vector of strings as input.

Output can be used with Darknet::load_neural_network().

Note

See the full description in the other Darknet::parse_arguments().

Since

2024-07-29

Todo:

confirm that this works just as well on ARM

Here is the call graph for this function:

parse_arguments() [2/2]

Darknet::Parms Darknet::parse_arguments	(	int	argc,
		char *	argv[]
	)

Parse common Darknet command-line parameters with the values from argc and argv in main().

Output can be used with Darknet::load_neural_network().

This function will attempt to identify the following:

• .cfg files (Darknet configuration)
• .names files (Darknet classes)
• .weights files (Darknet weights)
• all other files or subdirectories
• any other non-file parameters

If the .names or .weights files were not found, then attempts will be made to locate a suitable file to use based on the given .cfg file. This means as long as the 3 Darknet files are named in a similar way, there is no need to specify all 3 files. The .cfg file is enough to find the neural network.

In addition, if no Darknet files were found but a "stem" was specified, then this stem will be used to attempt and find all the necessary files. For example, if the following files exist:

• animals.cfg
• animals.names
• animals_best.weights

These files will be found if this function is called with the parameter "anim", since all the files begin with that stem.

Given the example files listed above, the following commands are interpreted the exact same way:

darknet_01_inference_images animals.cfg animals.names animals_best.weights dog.jpg
darknet_01_inference_images animals.names dog.jpg animals_best.weights animals.cfg
darknet_01_inference_images animals.cfg animals.names dog.jpg
darknet_01_inference_images animals.cfg dog.jpg
darknet_01_inference_images ani dog.jpg

Since

2024-07-29

Here is the call graph for this function:

Here is the caller graph for this function:

predict() [1/3]

Darknet::Predictions Darknet::predict	(	const Darknet::NetworkPtr	ptr,
		const cv::Mat &	mat
	)

Get Darknet to look at the given image or video frame and return all predictions.

This is similar to the other Darknet::predict() that takes a Darknet::Image object as input.

OpenCV cv::Mat images (and video frames) are typically stored in BGR format, not RGB. This function expects the images to be in the usual BGR format for 3-channel networks.

Since

2024-07-24

Here is the call graph for this function:

Here is the caller graph for this function:

predict() [2/3]

Darknet::Predictions Darknet::predict	(	const Darknet::NetworkPtr	ptr,
		const std::filesystem::path &	image_filename
	)

Get Darknet to look at the given image and return all predictions.

The image must be in a format supported by OpenCV, such as JPG or PNG.

Since

2024-07-25

Here is the call graph for this function:

predict() [3/3]

Darknet::Predictions Darknet::predict	(	const Darknet::NetworkPtr	ptr,
		Darknet::Image &	img,
		cv::Size	original_image_size = cv::Size(0, 0)
	)

Get Darknet to look at the given image or video frame and return all predictions.

The provided image must be in Darknet's RGB image format. This is similar to the other predict() that takes a cv::Mat object as input.

Note

If the original image size is known, it is best to pass it in so the bounding boxes can be scaled to those dimensions.

Since

2024-08-02

todo pass net by ref or pointer, not copy constructor!

Here is the call graph for this function:

predict_and_annotate()

Darknet::Predictions Darknet::predict_and_annotate	(	const Darknet::NetworkPtr	ptr,
		cv::Mat	mat
	)

Combination of Darknet::predict() and Darknet::annotate().

Remember to clone mat prior to calling predict_and_annotate() if you need to keep a copy of the original image.

Since

2024-07-30

Here is the call graph for this function:

Here is the caller graph for this function:

quantize_image()

void Darknet::quantize_image

(

Darknet::Image &

im

)

Note

Function currently seems to be unused.

random_augment_image()

Darknet::Image Darknet::random_augment_image	(	const Darknet::Image &	im,
		const float	angle,
		float	aspect,
		const int	low,
		const int	high,
		const int	size
	)

Apply a bunch of random augmentations.

Note

Function currently seems to be unused.

Here is the call graph for this function:

Here is the caller graph for this function:

random_crop_image()

Darknet::Image Darknet::random_crop_image	(	const Darknet::Image &	im,
		const int	w,
		const int	h
	)

Similar to Darknet::crop_image() but the dx and dy are random values.

Here is the call graph for this function:

random_distort_image()

void Darknet::random_distort_image	(	Darknet::Image &	im,
		float	hue,
		float	saturation,
		float	exposure
	)

Todo:

COLOR - HSV no beuno

Here is the call graph for this function:

read_tree()

Darknet::Tree * Darknet::read_tree

(

const char *

filename

)

Here is the call graph for this function:

Here is the caller graph for this function:

resize_image()

Darknet::Image Darknet::resize_image	(	const Darknet::Image &	im,
		int	w,
		int	h
	)

Here is the call graph for this function:

Here is the caller graph for this function:

resize_keeping_aspect_ratio()

cv::Mat Darknet::resize_keeping_aspect_ratio	(	cv::Mat &	mat,
		cv::Size	desired_size,
		const cv::InterpolationFlags	method = cv::InterpolationFlags::INTER_NEAREST
	)

Resize the image as close as we can to the given size, but keep the aspect ratio the same as the original image.

This method will modify the image that is passed in, so clone it beforehand if you need a copy of the original.

Several notes in regards to the resize method:

• cv::InterpolationFlags::INTER_NEAREST is the fastest resize method, but the quality is poor.
• cv::InterpolationFlags::INTER_AREA is good when shrinking an image.
• cv::InterpolationFlags::INTER_CUBIC is good when growing an image
• cv::InterpolationFlags::INTER_LINEAR is similar to INTER_CUBIC, but faster

Since

2024-09-05

Here is the caller graph for this function:

resize_max()

Darknet::Image Darknet::resize_max	(	const Darknet::Image &	im,
		int	max
	)

Note

Function currently seems to be unused.

Here is the call graph for this function:

resize_min()

Darknet::Image Darknet::resize_min	(	const Darknet::Image &	im,
		int	min
	)

Note

Function currently seems to be unused.

Here is the call graph for this function:

rgb_image_to_bgr_mat()

cv::Mat Darknet::rgb_image_to_bgr_mat

(

const Darknet::Image &

img

)

Similar to the original image_to_mat(), but with 2 differences:

• the output image is in the "natural" OpenCV BGR format,
• this function uses very efficient OpenCV techniques to convert the Darknet::Image to cv::Mat which makes it much faster.

See also

Darknet::bgr_mat_to_rgb_image()

Darknet::image_to_mat()

Darknet::mat_to_image()

Since

2024-08-23

rgb_to_hsv()

void Darknet::rgb_to_hsv

(

Darknet::Image &

im

)

Note

Function currently seems to be unused.

Todo:

COLOR - cannot do HSV if channels > 3

Here is the call graph for this function:

Here is the caller graph for this function:

rgbgr_image()

void Darknet::rgbgr_image

(

Darknet::Image &

im

)

Do the equivalent of OpenCV's cv::COLOR_BGR2RGB to swap red and blue floats.

Here is the caller graph for this function:

rotate_crop_image()

Darknet::Image Darknet::rotate_crop_image	(	const Darknet::Image &	im,
		float	rad,
		float	s,
		int	w,
		int	h,
		float	dx,
		float	dy,
		float	aspect
	)

Here is the call graph for this function:

Here is the caller graph for this function:

rotate_image()

Darknet::Image Darknet::rotate_image	(	const Darknet::Image &	im,
		float	rad
	)

Rotate image.

Note

Function currently seems to be unused.

Here is the call graph for this function:

Here is the caller graph for this function:

rotate_image_cw()

void Darknet::rotate_image_cw	(	Darknet::Image &	im,
		int	times
	)

Rotate image clockwise?

Note

Function currently seems to be unused.

run_image_loading_control_thread()

void Darknet::run_image_loading_control_thread

(

load_args

args

)

This runs as a std::thread.

It is started by the main thread during training and ensures the data-loading threads are running. This starts the thread that controls all of the permanent image loading threads.

This was originally called load_threads() and used pthread, but has since been re-written to use C++11.

See also

stop_image_loading_threads()

Since

2024-03-31

Here is the call graph for this function:

Here is the caller graph for this function:

save_image()

void Darknet::save_image	(	const Darknet::Image &	p,
		const char *	name
	)

Here is the call graph for this function:

Here is the caller graph for this function:

save_image_jpg()

void Darknet::save_image_jpg	(	const Darknet::Image &	im,
		const char *	name
	)

Todo:

merge with save_mat_jpg()

Here is the call graph for this function:

Here is the caller graph for this function:

save_image_png()

void Darknet::save_image_png	(	const Darknet::Image &	im,
		const char *	name
	)

Todo:

merge with save_mat_png()

Here is the call graph for this function:

scale_image()

void Darknet::scale_image	(	Darknet::Image &	m,
		const float	s
	)

Scale the RGB values in an image by the given amount.

Here is the caller graph for this function:

scale_image_channel()

void Darknet::scale_image_channel	(	Darknet::Image &	im,
		int	c,
		float	v
	)

Here is the call graph for this function:

Here is the caller graph for this function:

set_annotation_draw_bb()

void Darknet::set_annotation_draw_bb	(	Darknet::NetworkPtr	ptr,
		const bool	toggle
	)

Determines if bounding boxes are drawn when calling either Darknet::annotate() or Darknet::predict_and_annotate().

The default is true.

Since

2024-07-30

set_annotation_draw_label()

void Darknet::set_annotation_draw_label	(	Darknet::NetworkPtr	ptr,
		const bool	toggle
	)

Determines if text labels are drawn above the bounding boxes when calling either Darknet::annotate() or Darknet::predict_and_annotate().

The default is true.

Since

2024-07-30

set_annotation_font()

void Darknet::set_annotation_font	(	Darknet::NetworkPtr	ptr,
		const cv::LineTypes	line_type,
		const cv::HersheyFonts	font_face,
		const int	font_thickness,
		const double	font_scale
	)

Set the font characteristics to use when drawing the bounding boxes and labels in either Darknet::annotate() or Darknet::predict_and_annotate().

Parameters

[in]	ptr	Neural network pointer obtained from Darknet::load_neural_network().
[in]	line_type	should be cv::LineTypes::LINE_4, cv::LineTypes::LINE_8, or cv::LineTypes::CV_LINE_AA. LINE_4 is the fastest but lower quality, while LINE_AA (anti-alias) is the slowest with highest quality. Default is LINE_4. Also see set_annotation_line_type() which modifies the same setting.
[in]	font_face	is the OpenCV built-in font to use. Default is cv::HersheyFonts::FONT_HERSHEY_PLAIN.
[in]	font_thickness	determines how thick the lines are drawn when the text is rendered. Default is 1.
[in]	font_scale	determines how large or small the text is rendered. For example, this could be set to 0.5 for extremely small text, and 1.75 for large text. Default is 1.0.

See also

Darknet::NetworkDetails::cv_line_type

Darknet::NetworkDetails::cv_font_face

Darknet::NetworkDetails::cv_font_thickness

Darknet::NetworkDetails::cv_font_scale

Since

2024-07-30

Here is the caller graph for this function:

set_annotation_line_type()

void Darknet::set_annotation_line_type	(	Darknet::NetworkPtr	ptr,
		const cv::LineTypes	line_type
	)

The OpenCV line type can impact performance.

Anti-aliased lines are expensive to draw. Possible options for line_type is cv::LineTypes::LINE_4, cv::LineTypes::LINE_8, or cv::LineTypes::CV_LINE_AA. LINE_4 is the fastest but lower quality, while LINE_AA (anti-alias) is the slowest with highest quality. Default is LINE_4.

This setting can also be modified with set_annotation_font().

Since

2024-09-03

Here is the caller graph for this function:

set_class_colours()

const Darknet::VScalars & Darknet::set_class_colours	(	Darknet::NetworkPtr	ptr,
		const Darknet::VScalars &	colours
	)

Set the colours to use when drawing annotations.

The colours are in OpenCV's usual BGR format, not RGB. So pure red for example is "{0, 0, 255}" while pure blue would be "{255, 0, 0}". The middle value is green.

Returns

the final colours used, same as if get_class_colours() had been called.

Since

2024-09-22

set_default_keypoints_files()

void Darknet::set_default_keypoints_files

(

Darknet::Parms &

parms

)

Add the default Darknet Keypoints neural network files to the given parms.

If parms doesn't seem to reference neural network files, then this function will automatically add the names "Darknet-Keypoints.cfg" and "Darknet-Keypoints.weights".

Since

2024-09-20

Here is the caller graph for this function:

set_detection_threshold()

void Darknet::set_detection_threshold	(	Darknet::NetworkPtr	ptr,
		float	threshold
	)

Detection threshold to use when Darknet::predict() is called.

Default is 0.25.

See also

Darknet::NetworkDetails::detection_threshold

Since

2024-07-24

Here is the caller graph for this function:

set_gpu_index()

void Darknet::set_gpu_index

(

int

idx

)

Set the GPU index to use.

This may be set to -1 to indicate no GPU has been selected, or may be set to a 0-based GPU. In normal situations, this must be set prior to calling Darknet::load_neural_network() where the GPU is usually initialized.

If set to -1 and Darknet was compiled with support for CUDA GPUs, then the GPU index will default to 0 when Darknet::load_neural_network() is called. If no CUDA GPU is detected, then the GPU index will be set to -1 and the CPU will be used instead.

Default is -1.

Since

2024-07-25

Here is the caller graph for this function:

set_non_maximal_suppression_threshold()

void Darknet::set_non_maximal_suppression_threshold	(	Darknet::NetworkPtr	ptr,
		float	threshold
	)

Non-maximal suppression threshold to use when Darknet::predict() is called.

Default is 0.45.

See also

Darknet::NetworkDetails::non_maximal_suppression_threshold

Since

2024-07-24

Here is the caller graph for this function:

set_rounded_corner_bounding_boxes()

void Darknet::set_rounded_corner_bounding_boxes	(	Darknet::NetworkPtr	ptr,
		const bool	toggle,
		const float	roundness
	)

This determines if annotations are drawn as circles or rounded rectangles in either Darknet::annotate() or Darknet::predict_and_annotate().

The defaul is to use square – not rounded – bounding boxes.

Parameters

[in]	ptr	Neural network pointer obtained from Darknet::load_neural_network().
[in]	toggle	Determines if rounded corners are used. The default is false in which case normal "square" bounding boxes are used.
[in]	roundness	Determines how large the rounded corners will appear. The value must be between 0.0 (small rounded corners) and 1.0 (large rounded corners). At the extreme of 1.0, the bounding box will appear as a circle. The default is 0.5, but will only take effect if rounded is also set to true.

Since

2024-07-30

Here is the caller graph for this function:

set_trace()

void Darknet::set_trace

(

const bool

flag

)

Set the Darknet::CfgAndState::is_trace flag.

When enabled, debug information will be sent to STDOUT. Default value is false.

Note

Enabling trace will also enable verbose.

See also

Darknet::set_verbose()

darknet_set_trace()

Since

2024-07-14

Here is the call graph for this function:

Here is the caller graph for this function:

set_verbose()

void Darknet::set_verbose

(

const bool

flag

)

Set the Darknet::CfgAndState::is_verbose flag.

When enabled, extra information will be sent to STDOUT. Default value is false.

Note

Disabling verbose will also disable trace.

See also

Darknet::set_trace()

darknet_set_verbose()

Since

2024-07-14

Here is the call graph for this function:

Here is the caller graph for this function:

show_image()

void Darknet::show_image	(	const Darknet::Image &	p,
		const char *	name
	)

Here is the call graph for this function:

Here is the caller graph for this function:

show_image_collapsed()

void Darknet::show_image_collapsed	(	const Darknet::Image &	p,
		const char *	name
	)

Here is the call graph for this function:

show_image_layers()

void Darknet::show_image_layers	(	const Darknet::Image &	p,
		const char *	name
	)

Here is the call graph for this function:

show_images()

void Darknet::show_images	(	Darknet::Image *	ims,
		int	n,
		const char *	window
	)

Here is the call graph for this function:

Here is the caller graph for this function:

show_version_info()

void Darknet::show_version_info

(

)

Display a few lines of text with some version information.

Since

2024-08-29

Here is the call graph for this function:

Here is the caller graph for this function:

skipped_classes() [1/2]

Darknet::SInt Darknet::skipped_classes

(

const Darknet::NetworkPtr

ptr

)

Return the set of classes which Darknet must ignore.

Default set is empty.

See also

Darknet::skipped_classes()

Darknet::clear_skipped_classes()

Darknet::add_skipped_class()

Darknet::del_skipped_class()

Since

2024-10-07

skipped_classes() [2/2]

Darknet::SInt Darknet::skipped_classes	(	Darknet::NetworkPtr	ptr,
		const SInt &	classes_to_skip
	)

Set the classes which Darknet must ignore, completely over-writing all previous values.

If you'd rather add a single class at a time, call add_skipped_class() which can be called repeatedly without overwriting previous settings.

See also

Darknet::skipped_classes()

Darknet::clear_skipped_classes()

Darknet::add_skipped_class()

Darknet::del_skipped_class()

Since

2024-10-07

stop_image_loading_threads()

void Darknet::stop_image_loading_threads

(

)

Stop and join the image loading threads started in Darknet::run_image_loading_control_thread().

This was originally called free_load_threads() and used pthread, but has since been re-written to use C++11.

Since

2024-04-02

Here is the caller graph for this function:

test_resize()

void Darknet::test_resize

(

char *

filename

)

Here is the call graph for this function:

Here is the caller graph for this function:

text_to_simple_label()

std::string Darknet::text_to_simple_label

(

std::string

txt

)

Convert the given text to plain alphanumeric ASCII string.

Remove whitespace, keep just alphanumeric and underscore. Good to use as a base for a filename.

Here is the caller graph for this function:

threshold_image()

Darknet::Image Darknet::threshold_image	(	const Darknet::Image &	im,
		float	thresh
	)

Here is the call graph for this function:

Here is the caller graph for this function:

to_string() [1/9]

std::string Darknet::to_string

(

const EActivation

activation

)

Convert between names and activation types.

Here is the call graph for this function:

to_string() [2/9]

std::string Darknet::to_string

(

const ECostType

type

)

Convert between names and cost types.

Here is the call graph for this function:

to_string() [3/9]

std::string Darknet::to_string

(

const EIoULoss

loss

)

Convert between names and IoU loss types.

Here is the call graph for this function:

to_string() [4/9]

std::string Darknet::to_string

(

const ELayerType

type

)

Convert between names and Darknet/YOLO layer types.

Here is the call graph for this function:

Here is the caller graph for this function:

to_string() [5/9]

std::string Darknet::to_string

(

const ELearningRatePolicy

policy

)

Convert between names and learning rate policies.

Here is the call graph for this function:

to_string() [6/9]

std::string Darknet::to_string

(

const ENMSKind

nms_kind

)

Convert between names and NMS kind.

Here is the call graph for this function:

to_string() [7/9]

std::string Darknet::to_string

(

const EWeightsNormalization

normalization

)

Convert between names and weights normalization.

Here is the call graph for this function:

to_string() [8/9]

std::string Darknet::to_string

(

const EWeightsType

type

)

Convert between names and weights types.

Here is the call graph for this function:

to_string() [9/9]

std::string Darknet::to_string

(

const EYoloPoint

type

)

Convert between names and YOLO point types.

Here is the call graph for this function:

translate_image()

void Darknet::translate_image	(	Darknet::Image	m,
		float	s
	)

trim() [1/2]

std::string Darknet::trim

(

const std::string &

str

)

Trim leading and trailing whitespace from the given string.

Here is the call graph for this function:

Here is the caller graph for this function:

trim() [2/2]

std::string & Darknet::trim

(

std::string &

str

)

Trim leading and trailing whitespace from the given string.

update_accuracy_in_new_charts()

void Darknet::update_accuracy_in_new_charts	(	const int	class_index,
		const float	accuracy
	)

Here is the call graph for this function:

Here is the caller graph for this function:

update_loss_in_new_charts()

void Darknet::update_loss_in_new_charts	(	const int	current_iteration,
		const float	loss,
		const float	seconds_remaining,
		const bool	dont_show
	)

Update the new C++ charts with the given loss and mAP% accuracy value. This is called at every iteration.

See also

Chart

Here is the call graph for this function:

Here is the caller graph for this function: