darknet Namespace Reference

Classes
class	BOX
class	DETECTION
class	DETNUMPAIR
class	IMAGE

Functions
	bbox2points (bbox)
	class_colors (names)
	decode_detection (detections)
	detect_image (network, class_names, image, thresh=.5, hier_thresh=.5, nms=.45)
	draw_boxes (detections, image, colors)
	load_network (config_file, data_file, weights, batch_size=1)
	network_dimensions (net)
	network_height (net)
	network_width (net)
	non_max_suppression_fast (detections, overlap_thresh)
	print_detections (detections, coordinates=False)
	remove_negatives (detections, class_names, num)
	remove_negatives_faster (detections, class_names, num)

Variables
	add_skipped_class = lib.darknet_add_skipped_class
	argtypes
	clear_skipped_classes = lib.darknet_clear_skipped_classes
	copy_image_from_bytes = lib.copy_image_from_bytes
	del_skipped_class = lib.darknet_del_skipped_class
	do_nms_obj = lib.do_nms_obj
	do_nms_sort = lib.do_nms_sort
	free_detections = lib.free_detections
	free_image = lib.free_image
	free_network_ptr = lib.free_network_ptr
	free_ptrs = lib.free_ptrs
	get_network_boxes = lib.get_network_boxes
	lib = CDLL(libpath, RTLD_GLOBAL)
str	libpath = "/usr/lib/libdarknet.so"
	load_image = lib.load_image_v2
	load_net = lib.load_network
	load_net_custom = lib.load_network_custom
	make_image = lib.make_image
	make_network_boxes = lib.make_network_boxes
	network_dimensions = lib.darknet_network_dimensions
	network_predict = lib.network_predict_ptr
	predict = lib.network_predict_ptr
	predict_image = lib.network_predict_image
	predict_image_letterbox = lib.network_predict_image_letterbox
	reset_rnn = lib.reset_rnn
	restype
	set_gpu = lib.cuda_set_device
	set_output_stream = lib.darknet_set_output_stream
	set_verbose = lib.darknet_set_verbose
	show_version_info = lib.darknet_show_version_info
	version_short = lib.darknet_version_short
	version_string = lib.darknet_version_string

Detailed Description

Python 3 wrapper for identifying objects in images

Running the script requires opencv-python to be installed, via either of these two options:

- sudo apt-get install python3-opencv
- pip3 install opencv-python

NOTE:  As of December 2023 while using Ubuntu 22.04, the apt-get install one was
better since the pip3 one would result in a known segfault when cv2.imshow() was called:
https://github.com/opencv/opencv-python/issues/794

Directly viewing or returning bounding-boxed images requires scikit-image to be installed:

- pip3 install scikit-image

See the example code (such as "example.py") which imports this module.

Function Documentation

bbox2points()

darknet.bbox2points

(

bbox

)

Convert bounding box from YOLO format to corner points in CV2 rectangle format.

Here is the caller graph for this function:

class_colors()

darknet.class_colors

(

names

)

Create a dict with a deterministic, yet 'random', BGR color for each class name.
This function uses a hash of the class name to generate consistent colors.

Here is the caller graph for this function:

decode_detection()

darknet.decode_detection

(

detections

)

Decode detection results and format confidence scores.

Args:
    detections: List of detected objects, each represented as (label, confidence, bbox).

Returns:
    List of decoded detection results with formatted confidence scores.

Here is the caller graph for this function:

detect_image()

darknet.detect_image	(		network,
			class_names,
			image,
			thresh = .5,
			hier_thresh = .5,
			nms = .45
	)

Returns a list with the highest confidence class and their bounding box.

Args:
    network: Darknet network.
    class_names: List of class names.
    image: Input image.
    thresh: Detection confidence threshold.
    hier_thresh: Hierarchical threshold.
    nms: Non-Maximum Suppression threshold.

Returns:
    List of detections with class name, confidence, and bounding box.

Here is the call graph for this function:

Here is the caller graph for this function:

draw_boxes()

darknet.draw_boxes	(		detections,
			image,
			colors
	)

Draw bounding boxes on an image.

Args:
    detections: List of detected objects, each represented as (label, confidence, bbox).
    image: Input image.
    colors: Dictionary of colors for each class label.

Returns:
    Image with bounding boxes drawn.

Here is the call graph for this function:

Here is the caller graph for this function:

load_network()

darknet.load_network	(		config_file,
			data_file,
			weights,
			batch_size = 1
	)

load model description and weights from config files
args:
    config_file (str): path to .cfg model file
    data_file (str): path to .data model file
    weights (str): path to weights
returns:
    network: trained model

network_dimensions()

darknet.network_dimensions

(

net

)

network_height()

darknet.network_height

(

net

)

Here is the caller graph for this function:

network_width()

darknet.network_width

(

net

)

Here is the caller graph for this function:

non_max_suppression_fast()

darknet.non_max_suppression_fast	(		detections,
			overlap_thresh
	)

Apply non-maximum suppression (NMS) to a list of detections.

Args:
    detections: List of detected objects, where each detection is represented as a tuple
                (label, confidence, (x, y, w, h)).
    overlap_thresh: Threshold for considering overlap between bounding boxes.

Returns:
    List of selected detections after NMS.

print_detections()

darknet.print_detections	(		detections,
			coordinates = False
	)

Print detected objects and their confidence scores.

Args:
    detections: List of detected objects, each represented as (label, confidence, bbox).
    coordinates: If True, also print bounding box coordinates.

Prints:
    Detected objects with or without bounding box coordinates.

Here is the caller graph for this function:

remove_negatives()

darknet.remove_negatives	(		detections,
			class_names,
			num
	)

Remove all classes with 0% confidence within the detection.

Args:
    detections: List of detections, where each detection is an object with confidence scores for classes.
    class_names: List of class names.
    num: Number of detections.

Returns:
    List of valid predictions with class name, confidence, and bounding box.

Here is the caller graph for this function:

remove_negatives_faster()

darknet.remove_negatives_faster	(		detections,
			class_names,
			num
	)

Faster version of remove_negatives (very useful when using yolo9000).

Args:
    detections: List of detections, where each detection is an object with confidence scores for classes.
    class_names: List of class names.
    num: Number of detections.

Returns:
    List of valid predictions with class name, confidence, and bounding box.

Variable Documentation

add_skipped_class

darknet.add_skipped_class = lib.darknet_add_skipped_class

argtypes

darknet.argtypes

clear_skipped_classes

darknet.clear_skipped_classes = lib.darknet_clear_skipped_classes

copy_image_from_bytes

darknet.copy_image_from_bytes = lib.copy_image_from_bytes

del_skipped_class

darknet.del_skipped_class = lib.darknet_del_skipped_class

do_nms_obj

darknet.do_nms_obj = lib.do_nms_obj

do_nms_sort

darknet.do_nms_sort = lib.do_nms_sort

free_detections

darknet.free_detections = lib.free_detections

free_image

darknet.free_image = lib.free_image

free_network_ptr

darknet.free_network_ptr = lib.free_network_ptr

free_ptrs

darknet.free_ptrs = lib.free_ptrs

get_network_boxes

darknet.get_network_boxes = lib.get_network_boxes

lib

darknet.lib = CDLL(libpath, RTLD_GLOBAL)

libpath

str darknet.libpath = "/usr/lib/libdarknet.so"

load_image

darknet.load_image = lib.load_image_v2

load_net

darknet.load_net = lib.load_network

load_net_custom

darknet.load_net_custom = lib.load_network_custom

make_image

darknet.make_image = lib.make_image

make_network_boxes

darknet.make_network_boxes = lib.make_network_boxes

network_dimensions

darknet.network_dimensions = lib.darknet_network_dimensions

network_predict

darknet.network_predict = lib.network_predict_ptr

predict

darknet.predict = lib.network_predict_ptr

predict_image

darknet.predict_image = lib.network_predict_image

predict_image_letterbox

darknet.predict_image_letterbox = lib.network_predict_image_letterbox

reset_rnn

darknet.reset_rnn = lib.reset_rnn

restype

darknet.restype

set_gpu

darknet.set_gpu = lib.cuda_set_device

set_output_stream

darknet.set_output_stream = lib.darknet_set_output_stream

set_verbose

darknet.set_verbose = lib.darknet_set_verbose

show_version_info

darknet.show_version_info = lib.darknet_show_version_info

version_short

darknet.version_short = lib.darknet_version_short

version_string

darknet.version_string = lib.darknet_version_string