Note

This is the documentation for the latest development branch and may refer to features that are not available in released versions. If you are looking for a specific release, use the drop-down menu on the left.

YOLO Module API Manual

Overview

This manual is intended to guide developers in using the YOLO module to deploy models trained and converted using YOLOv5, YOLOv8, YOLO11, and YOLO26, helping users quickly interface with the YOLO source code and run trained models on the K230. For YOLO usage examples, refer to the documentation: yolo_battle

The task types supported by each class are as follows:

Class	classify	detect	segment	obb (rotated object detection)	pose (keypoint detection)
YOLOv5	✅	✅	✅	❌	❌
YOLOv8	✅	✅	✅	✅	✅
YOLO11	✅	✅	✅	✅	✅
YOLO26	✅	✅	✅	✅	✅

The corresponding examples are located in the development board’s CanMV/sdcard/examples/20-YOLO-Module-Examples directory, and each task type provides two scripts: *_image.py (single image inference) and *_video.py (video stream inference).

API Introduction

YOLOv5 Class

Constructor

Description

The constructor of the encapsulated YOLOv5 module initializes the YOLOv5 type to obtain a YOLOv5 instance.

Syntax

from libs.YOLO import YOLOv5

yolo=YOLOv5(task_type="classify",mode="image",kmodel_path="yolov5_det.kmodel",labels=["apple","banana","orange"],rgb888p_size=[1280,720],model_input_size=[320,320],conf_thresh=0.5,nms_thresh=0.25,max_boxes_num=50,debug_mode=0)

Parameters

Parameter Name	Description	Notes	Type
task_type	Task type	Supports three types of tasks, options are ‘classify’/’detect’/’segment’;	str
mode	Inference mode	Supports two inference modes, options are ‘image’/’video’, ‘image’ means inferring an image, ‘video’ means inferring the real-time video stream captured by the camera;	str
kmodel_path	kmodel path	The path of the kmodel copied to the development board;	str
labels	Class label list	Label names for different classes;	list[str]
rgb888p_size	Inference frame resolution	The resolution of the current inference frame, such as [1920,1080], [1280,720], [640,640];	list[int]
model_input_size	Model input resolution	The input resolution when the YOLOv5 model was trained, such as [224,224], [320,320], [640,640];	list[int]
display_size	Display resolution	Set when the inference mode is ‘video’, supports hdmi([1920,1080]) and lcd([800,480]);	list[int]
conf_thresh	Confidence threshold	Class confidence threshold for classification tasks, object confidence threshold for detection and segmentation tasks, such as 0.5;	float [0~1]
nms_thresh	nms threshold	Non-maximum suppression threshold, required for detection and segmentation tasks;	float [0~1]
mask_thresh	mask threshold	The binarization threshold for segmenting objects within the detection box in segmentation tasks;	float [0~1]
max_boxes_num	Maximum number of detection boxes	The maximum number of detection boxes allowed to be returned in one frame;	int
debug_mode	Debug mode	Whether the timing function is enabled, options are 0/1, 0 for no timing, 1 for timing;	int [0/1]

Return Value

Return Value	Description
YOLOv5	YOLOv5 instance

config_preprocess

Description

YOLOv5 preprocessing configuration function.

Syntax

yolo.config_preprocess()

Parameters

Parameter Name	Description	Input / Output	Notes
None

Return Value

Return Value	Description
None

run

Description

Infer one frame of image, and return the inference result for use by the draw_result method. The classification task returns the class index and score, the detection task returns a list of detection box positions, scores, and class indices, and the segmentation task returns the mask results and a list of detection box positions, scores, and class indices.

Syntax

res=yolo.run(img)

Parameters

Parameter Name	Description	Input / Output	Notes
img	The image to be inferred in the format of ulab.numpy.ndarray, or a frame of image obtained from the video stream via get_frame	Input

Return Value

Return Value	Description
res	Model post-processing result, which differs depending on the task. The classification task returns the class index and score, the detection task returns a list of detection box positions, scores, and class indices, and the segmentation task returns the mask results and a list of detection box positions, scores, and class indices.

draw_result

Description

Draw the YOLOv5 inference result on the screen or image.

Syntax

yolo.draw_result(res,img_ori)

Parameters

Parameter Name	Description	Input / Output	Notes
res	The inference result of YOLOv5	Input
img_ori	The Image instance to be drawn	Input	From read_img or pl.osd_img

Return Value

Return Value	Description
None

deinit

Description

Release the native resources occupied by the YOLOv5 instance, such as the kmodel, AI2D, and the mask cache for segmentation tasks. This must be called when inference ends or when switching models; otherwise, repeatedly creating instances will exhaust the nncase memory pool.

Syntax

yolo.deinit()

Parameters

Parameter Name	Description	Input / Output	Notes
None

Return Value

Return Value	Description
None	Safe to call repeatedly.

Tip

deinit() internally completes gc.collect() and nncase_runtime.shrink_memory_pool(), which returns memory to the nncase memory pool, so there is no need to manually call these again when subsequently loading another kmodel in the same script. Repeated calls to deinit() are safe; if a native resource fails to be released, the instance will retain it for the next retry and throw an exception.

Example Program

The following is an example program for the YOLOv5 detection task:

from libs.YOLO import YOLOv5
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is only an example. For custom scenarios, please modify it to your own test image, model path, label names, and model input size
    img_path="/sdcard/examples/utils/test_fruit.jpg"
    kmodel_path="/sdcard/examples/kmodel/fruit_det_yolov5n_320.kmodel"
    labels = ["apple","banana","orange"]
    model_input_size=[320,320]

    confidence_threshold = 0.5
    nms_threshold=0.45
    img,img_ori=read_image(img_path)
    rgb888p_size=[img.shape[2],img.shape[1]]
    # Initialize the YOLOv5 instance
    yolo=YOLOv5(task_type="detect",mode="image",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,conf_thresh=confidence_threshold,nms_thresh=nms_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    res=yolo.run(img)
    yolo.draw_result(res,img_ori)
    yolo.deinit()
    gc.collect()

The above code shows the code for performing image inference using YOLOv5.

from libs.PipeLine import PipeLine
from libs.YOLO import YOLOv5
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is only an example. For custom scenarios, please modify it to your own model path, label names, and model input size
    kmodel_path="/sdcard/examples/kmodel/fruit_det_yolov5n_320.kmodel"
    labels = ["apple","banana","orange"]
    model_input_size=[320,320]

    # Display mode, "auto" automatically selects the screen driver according to the development board model; you can also explicitly specify hdmi/lcd/st7701, etc. Refer to the PipeLine manual for valid values
    display_mode="auto"
    # None means using the default resolution of the screen driver; fill in [width, height] when specifying
    display_size=None
    rgb888p_size=[640,360]
    confidence_threshold = 0.8
    nms_threshold=0.45
    pl=PipeLine(rgb888p_size=rgb888p_size,display_mode=display_mode,display_size=display_size)
    pl.create()
    # Get the actual display resolution after create
    display_size=pl.get_display_size()
    # Initialize the YOLOv5 instance
    yolo=YOLOv5(task_type="detect",mode="video",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,display_size=display_size,conf_thresh=confidence_threshold,nms_thresh=nms_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    while True:
        with ScopedTiming("total",1):
            img=pl.get_frame()
            res=yolo.run(img)
            yolo.draw_result(res,pl.osd_img)
            pl.show_image()
            gc.collect()
    yolo.deinit()
    pl.destroy()

The above code shows the code for performing video inference using YOLOv5.

YOLOv8 Class

Constructor

Description

Encapsulated YOLOv8 module constructor, initializes the YOLOv8 type to get a YOLOv8 instance.

Syntax

from libs.YOLO import YOLOv8

yolo=YOLOv8(task_type="classify",mode="image",kmodel_path="yolov8_det.kmodel",labels=["apple","banana","orange"],rgb888p_size=[1280,720],model_input_size=[320,320],conf_thresh=0.5,nms_thresh=0.25,max_boxes_num=50,debug_mode=0)

Parameters

Parameter Name	Description	Instructions	Type
task_type	Task Type	Supports five types of tasks, options are ‘classify’/’detect’/’segment’/’obb’/’pose’;	str
mode	Inference Mode	Supports two inference modes, options are ‘image’/’video’, ‘image’ means inference on an image, ‘video’ means inference on a real-time video stream captured by the camera;	str
kmodel_path	kmodel Path	Path of the kmodel copied to the development board;	str
labels	Category Label List	Names of labels for different categories;	list[str]
rgb888p_size	Inference Frame Resolution	Resolution of the current frame to infer, such as [1920,1080], [1280,720], [640,640];	list[int]
model_input_size	Model Input Resolution	Input resolution used during YOLOv8 model training, such as [224,224], [320,320], [640,640];	list[int]
display_size	Display Resolution	Set when the inference mode is ‘video’, supports hdmi([1920,1080]) and lcd([800,480]);	list[int]
conf_thresh	Confidence Threshold	Category confidence threshold for classification tasks, target confidence threshold for detection and segmentation tasks, such as 0.5;	float [0~1]
nms_thresh	nms Threshold	Non-maximum suppression threshold, required for detection and segmentation tasks;	float [0~1]
mask_thresh	mask Threshold	Binarization threshold for segmenting objects in the detection box in segmentation tasks;	float [0~1]
max_boxes_num	Max Detection Boxes	Max number of detection boxes allowed to be returned in one frame;	int
kp_num	Number of Keypoints	Number of keypoints in the keypoint detection task;	int
kp_dim	Keypoint Dimension	Dimension of keypoints in the keypoint detection task, only 2 and 3 are supported, determined by the trained model;	int【2/3】
debug_mode	Debug Mode	Whether the timing function is enabled, options 0/1, 0 means no timing, 1 means timing;	int [0/1]

Return Value

Return Value	Description
YOLOv8	YOLOv8 instance

config_preprocess

Description

YOLOv8 preprocessing configuration function.

Syntax

yolo.config_preprocess()

Parameters

Parameter Name	Description	Input / Output	Instructions
None

Return Value

Return Value	Description
None

run

Description

Infer one frame of image, return the inference result for use by the draw_result method. The classification task returns the category index and score, the detection task returns the list of detection box positions, scores, and category indices, the segmentation task returns the mask results and the list of detection box positions, scores, and category indices, the rotated object detection returns the list of four-point coordinates of the rotated detection box, and the keypoint detection returns the detection box positions, category indices, scores, and keypoint list.

Syntax

res=yolo.run(img)

Parameters

Parameter Name	Description	Input / Output	Instructions
img	An image to be inferred in the format of ulab.numpy.ndarray, or a frame of image obtained from the video stream via get_frame	Input

Return Value

Return Value

Description

res

Model post-processing result, different tasks return different values. The classification task returns the category index and score, the detection task returns the list of detection box positions, scores, and category indices, the segmentation task returns the mask results and the list of detection box positions, scores, and category indices, the rotated object detection returns the list of four-point coordinates of the rotated detection box, and the keypoint detection returns the detection box positions, category indices, scores, and keypoint list.

draw_result

Description

Draw the YOLOv8 inference result on the screen or image.

Syntax

yolo.draw_result(res,img_ori)

Parameters

Parameter Name	Description	Input / Output	Instructions
res	Inference result of YOLOv8	Input
img_ori	Image instance to be drawn	Input	From read_img or pl.osd_img

Return Value

Return Value	Description
None

deinit

Description

Release the native resources occupied by the YOLOv8 instance, such as kmodel, AI2D, and the mask cache for segmentation tasks. Must be called when inference ends or when switching models, otherwise repeatedly creating instances will exhaust the nncase memory pool.

Syntax

yolo.deinit()

Parameters

Parameter Name	Description	Input / Output	Instructions
None

Return Value

Return Value	Description
None	Safe to call repeatedly.

Tip

deinit() internally completes gc.collect() and nncase_runtime.shrink_memory_pool(), which will return memory to the nncase memory pool, so there is no need to manually call again when loading another kmodel in the same script. Repeated calls to deinit() are safe; if a certain native resource release fails, the instance will keep it for the next retry and throw an exception.

Example Program

The following gives an example program for the YOLOv8 classification task:

from libs.YOLO import YOLOv8
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is just an example. For custom scenarios, please modify to your own test image, model path, label name, and model input size
    img_path="/sdcard/examples/utils/test_fruit.jpg"
    kmodel_path="/sdcard/examples/kmodel/fruit_det_yolov8n_320.kmodel"
    labels = ["apple","banana","orange"]
    model_input_size=[320,320]

    confidence_threshold = 0.5
    nms_threshold=0.45
    img,img_ori=read_image(img_path)
    rgb888p_size=[img.shape[2],img.shape[1]]
    # Initialize YOLOv8 instance
    yolo=YOLOv8(task_type="detect",mode="image",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,conf_thresh=confidence_threshold,nms_thresh=nms_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    res=yolo.run(img)
    yolo.draw_result(res,img_ori)
    yolo.deinit()
    gc.collect()

The above code shows the code for image inference using YOLOv8.

from libs.PipeLine import PipeLine
from libs.YOLO import YOLOv8
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is just an example. For custom scenarios, please modify to your own model path, label name, and model input size
    kmodel_path="/sdcard/examples/kmodel/fruit_det_yolov8n_320.kmodel"
    labels = ["apple","banana","orange"]
    model_input_size=[320,320]

    # Display mode, "auto" automatically selects the screen driver according to the development board model; you can also explicitly specify hdmi/lcd/st7701, etc., see the PipeLine manual for values
    display_mode="auto"
    # None means using the default resolution of the screen driver; fill in [width, height] when you need to specify
    display_size=None
    rgb888p_size=[640,360]
    confidence_threshold = 0.5
    nms_threshold=0.45
    # Initialize PipeLine
    pl=PipeLine(rgb888p_size=rgb888p_size,display_mode=display_mode,display_size=display_size)
    pl.create()
    # Get the actual display resolution after create
    display_size=pl.get_display_size()
    # Initialize YOLOv8 instance
    yolo=YOLOv8(task_type="detect",mode="video",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,display_size=display_size,conf_thresh=confidence_threshold,nms_thresh=nms_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    while True:
        with ScopedTiming("total",1):
            # Infer frame by frame
            img=pl.get_frame()
            res=yolo.run(img)
            yolo.draw_result(res,pl.osd_img)
            pl.show_image()
            gc.collect()
    yolo.deinit()
    pl.destroy()

The above code shows the code for video inference using YOLOv8.

YOLO11 Class

Constructor

Description

Encapsulated YOLO11 module constructor. Initializes the YOLO11 type to obtain a YOLO11 instance.

Syntax

from libs.YOLO import YOLO11

yolo=YOLO11(task_type="segment",mode="image",kmodel_path="yolo11_det.kmodel",labels=["apple","banana","orange"],rgb888p_size=[1280,720],model_input_size=[320,320],conf_thresh=0.5,nms_thresh=0.25,mask_thresh=0.5,max_boxes_num=50,debug_mode=0)

Parameters

Parameter Name	Description	Notes	Type
task_type	Task type	Supports five task types, with optional values ‘classify’/’detect’/’segment’/’obb’/’pose’;	str
mode	Inference mode	Supports two inference modes, with optional values ‘image’/’video’; ‘image’ indicates inferring images, ‘video’ indicates inferring real-time video streams captured by the camera;	str
kmodel_path	kmodel path	Path of the kmodel copied to the development board;	str
labels	Class label list	Label names for different classes;	list[str]
rgb888p_size	Inference frame resolution	Resolution of the current frame being inferred, e.g., [1920,1080], [1280,720], [640,640];	list[int]
model_input_size	Model input resolution	Input resolution when the YOLO11 model was trained, e.g., [224,224], [320,320], [640,640];	list[int]
display_size	Display resolution	Set when the inference mode is ‘video’; supports hdmi([1920,1080]) and lcd([800,480]);	list[int]
conf_thresh	Confidence threshold	Class confidence threshold for classification tasks, object confidence threshold for detection and segmentation tasks, e.g., 0.5;	float [0~1]
nms_thresh	nms threshold	Non-maximum suppression threshold; required for detection and segmentation tasks;	float [0~1]
mask_thresh	mask threshold	Binarization threshold for segmenting the object inside the detection box in segmentation tasks;	float [0~1]
max_boxes_num	Maximum number of detection boxes	Maximum number of detection boxes allowed to be returned in one frame;	int
kp_num	Number of keypoints	Number of keypoints in a keypoint detection task;	int
kp_dim	Keypoint dimension	Dimension of keypoints in a keypoint detection task; only 2 and 3 are supported, determined by the trained model;	int【2/3】
debug_mode	Debug mode	Whether the timing function is enabled; optional values 0/1, 0 means no timing, 1 means timing;	int [0/1]

Return Value

Return Value	Description
YOLO11	YOLO11 instance

config_preprocess

Description

YOLO11 preprocessing configuration function.

Syntax

yolo.config_preprocess()

Parameters

Parameter Name	Description	Input / Output	Notes
None

Return Value

Return Value	Description
None

run

Description

Infer one frame of image and return the inference result for use by the draw_result method. For classification tasks, it returns the class index and score; for detection tasks, it returns a list of detection box positions, scores, and class indices; for segmentation tasks, it returns the mask results along with a list of detection box positions, scores, and class indices; for oriented object detection, it returns a list of the four point coordinates of the rotated detection box; for keypoint detection, it returns the detection box positions, class indices, scores, and keypoints list.

Syntax

res=yolo.run(img)

Parameters

Parameter Name	Description	Input / Output	Notes
img	Image to be inferred in the format of ulab.numpy.ndarray, or a frame obtained from the video stream via get_frame	Input

Return Value

Return Value

Description

res

Model post-processing result; different tasks return different values. For classification tasks, it returns the class index and score; for detection tasks, it returns a list of detection box positions, scores, and class indices; for segmentation tasks, it returns the mask results along with a list of detection box positions, scores, and class indices; for oriented object detection, it returns a list of the four point coordinates of the rotated detection box; for keypoint detection, it returns the detection box positions, class indices, scores, and keypoints list.

draw_result

Description

Draw the YOLO11 inference result on the screen or image.

Syntax

yolo.draw_result(res,img_ori)

Parameters

Parameter Name	Description	Input / Output	Notes
res	Inference result of YOLO11	Input
img_ori	Image instance to be drawn	Input	Comes from read_img or pl.osd_img

Return Value

Return Value	Description
None

deinit

Description

Releases the native resources occupied by the YOLO11 instance, such as the kmodel, AI2D, and the mask cache of segmentation tasks. Must be called when inference ends or when switching models; otherwise, repeatedly creating instances will exhaust the nncase memory pool.

Syntax

yolo.deinit()

Parameters

Parameter Name	Description	Input / Output	Notes
None

Return Value

Return Value	Description
None	Safe to call repeatedly.

Tip

deinit() internally completes gc.collect() and nncase_runtime.shrink_memory_pool(), which returns memory to the nncase memory pool. Therefore, there is no need to manually call these when loading another kmodel subsequently in the same script. Repeatedly calling deinit() is safe; if the release of a native resource fails, the instance will retain it for the next retry and raise an exception.

Example Program

The following gives an example program for the YOLO11 segmentation task:

from libs.YOLO import YOLO11
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is only an example; please modify it to your own test image, model path, label names, and model input size for your custom scenario
    img_path="/sdcard/examples/utils/test_fruit.jpg"
    kmodel_path="/sdcard/examples/kmodel/fruit_det_yolo11n_320.kmodel"
    labels = ["apple","banana","orange"]
    model_input_size=[320,320]

    confidence_threshold = 0.5
    nms_threshold=0.45
    img,img_ori=read_image(img_path)
    rgb888p_size=[img.shape[2],img.shape[1]]
    # Initialize the YOLO11 instance
    yolo=YOLO11(task_type="detect",mode="image",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,conf_thresh=confidence_threshold,nms_thresh=nms_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    res=yolo.run(img)
    yolo.draw_result(res,img_ori)
    yolo.deinit()
    gc.collect()

The code above gives an example of using YOLO11 for image inference.

from libs.PipeLine import PipeLine
from libs.YOLO import YOLO11
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is only an example; please modify it to your own model path, label names, and model input size for your custom scenario
    kmodel_path="/sdcard/examples/kmodel/fruit_det_yolo11n_320.kmodel"
    labels = ["apple","banana","orange"]
    model_input_size=[320,320]

    # Display mode, "auto" automatically selects the screen driver according to the development board model; can also explicitly specify hdmi/lcd/st7701, etc., see the PipeLine manual for valid values
    display_mode="auto"
    # None means using the default resolution of the screen driver; specify [width, height] if needed
    display_size=None
    rgb888p_size=[640,360]
    confidence_threshold = 0.5
    nms_threshold=0.45
    # Initialize PipeLine
    pl=PipeLine(rgb888p_size=rgb888p_size,display_mode=display_mode,display_size=display_size)
    pl.create()
    # Get the actual display resolution after create()
    display_size=pl.get_display_size()
    # Initialize the YOLO11 instance
    yolo=YOLO11(task_type="detect",mode="video",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,display_size=display_size,conf_thresh=confidence_threshold,nms_thresh=nms_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    while True:
        with ScopedTiming("total",1):
            # Frame-by-frame inference
            img=pl.get_frame()
            res=yolo.run(img)
            yolo.draw_result(res,pl.osd_img)
            pl.show_image()
            gc.collect()
    yolo.deinit()
    pl.destroy()

The code above gives an example of using YOLO11 for video inference.

YOLO26 Class

Constructor

Description

Constructor of the encapsulated YOLO26 module, initializes the YOLO26 type to obtain a YOLO26 instance.

Syntax

from libs.YOLO import YOLO26

yolo=YOLO26(task_type="pose",mode="image",kmodel_path="yolo26n_pose.kmodel",labels=["people"],rgb888p_size=[1280,720],model_input_size=[320,320],conf_thresh=0.5,mask_thresh=0.5,kp_num=17,kp_dim=3,max_boxes_num=50,debug_mode=0)

Parameters

Parameter Name	Description	Specification	Type
task_type	Task Type	Supports five types of tasks, options are ‘classify’/’detect’/’segment’/’obb’/’pose’;	str
mode	Inference Mode	Supports two inference modes, options are ‘image’/’video’, ‘image’ means inferring an image, ‘video’ means inferring the real-time video stream captured by the camera;	str
kmodel_path	kmodel Path	kmodel path copied to the development board;	str
labels	Category Labels List	Label names for different categories;	list[str]
rgb888p_size	Inference Frame Resolution	Resolution of the current inference frame, such as [1920,1080], [1280,720], [640,640];	list[int]
model_input_size	Model Input Resolution	Input resolution when training YOLO11 model, such as [224,224], [320,320], [640,640];	list[int]
display_size	Display Resolution	Set when the inference mode is ‘video’, supports hdmi([1920,1080]) and lcd([800,480]);	list[int]
conf_thresh	Confidence Threshold	Category confidence threshold for classification tasks, target confidence threshold for detection and segmentation tasks, such as 0.5;	float [0~1]
mask_thresh	Mask Threshold	Binarization threshold for segmenting objects in detection boxes in segmentation tasks;	float [0~1]
max_boxes_num	Maximum Detection Box Count	Maximum number of detection boxes allowed to be returned in one frame;	int
kp_num	Number of Keypoints	Number of keypoints in keypoint detection tasks;	int
kp_dim	Keypoint Dimension	Dimension of keypoints in keypoint detection tasks, only 2 and 3 are supported, determined by the training model;	int【2/3】
debug_mode	Debug Mode	Whether the timing function takes effect, options 0/1, 0 means no timing, 1 means timing;	int [0/1]

Note: The YOLO26 module no longer includes the nms_thresh parameter because the YOLO26 model directly outputs results with semantic information.

Return Value

Return Value	Description
YOLO26	YOLO26 instance

config_preprocess

Description

YOLO26 preprocessing configuration function.

Syntax

yolo.config_preprocess()

Parameters

Parameter Name	Description	Input / Output	Notes
None

Return Value

Return Value	Description
None

run

Description

Infer one frame of image and return the inference result for use by the draw_result method. Classification tasks return category index and score, detection tasks return a list of detection box positions, scores, and category indices, segmentation tasks return mask results and a list of detection box positions, scores, and category indices, rotated object detection returns a list of the four point coordinates of the rotated detection box, and keypoint detection returns a list of detection box positions, category indices, scores, and keypoints.

Syntax

res=yolo.run(img)

Parameters

Parameter Name	Description	Input / Output	Notes
img	Image to be inferred in the format of ulab.numpy.ndarray, or a frame obtained from the video stream through get_frame	Input

Return Value

Return Value

Description

res

Model post-processing results, different return values for different tasks. Classification tasks return category index and score, detection tasks return a list of detection box positions, scores, and category indices, segmentation tasks return mask results and a list of detection box positions, scores, and category indices, rotated object detection returns a list of the four point coordinates of the rotated detection box, and keypoint detection returns a list of detection box positions, category indices, scores, and keypoints.

draw_result

Description

Draws YOLO26 inference results on the screen or image.

Syntax

yolo.draw_result(res,img_ori)

Parameters

Parameter Name	Description	Input / Output	Notes
res	YOLO26 inference result	Input
img_ori	Image instance to be drawn	Input	From read_img or pl.osd_img

Return Value

Return Value	Description
None

deinit

Description

Releases native resources such as the kmodel, AI2D, and mask cache for segmentation tasks occupied by the YOLO26 instance. Must be called after inference is finished or when switching models; otherwise, repeatedly creating instances will exhaust the nncase memory pool.

Syntax

yolo.deinit()

Parameters

Parameter Name	Description	Input / Output	Notes
None

Return Value

Return Value	Description
None	Safe to call repeatedly.

Tip

deinit() internally has completed gc.collect() and nncase_runtime.shrink_memory_pool(), which will return memory to the nncase memory pool, so there is no need to manually call it again when subsequently loading another kmodel in the same script. Repeatedly calling deinit() is safe; if a certain native resource fails to be released, the instance will retain it for the next retry and throw an exception.

Example Programs

The following is an example program for the YOLO26 segmentation task:

from libs.YOLO import YOLO26
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is only an example, please modify it to your own test image, model path, label name, model input size, number of keypoints, and keypoint dimension for custom scenarios
    img_path="/sdcard/examples/utils/test_pose.jpg"
    kmodel_path="/sdcard/examples/kmodel/yolo26n-pose.kmodel"
    labels = ['person']
    model_input_size=[320,320]
    kp_num=17
    kp_dim=3

    confidence_threshold = 0.5
    img,img_ori=read_image(img_path)
    rgb888p_size=[img.shape[2],img.shape[1]]
    # Initialize YOLO26 model (YOLO26 has no nms_thresh parameter)
    yolo=YOLO26(task_type="pose",mode="image",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,kp_num=kp_num,kp_dim=kp_dim,conf_thresh=confidence_threshold,max_boxes_num=100,debug_mode=0)
    yolo.config_preprocess()
    res=yolo.run(img)
    print(res)
    yolo.draw_result(res,img_ori)
    yolo.deinit()
    gc.collect()

The above code shows the code for using YOLO26 for image inference.

from libs.PipeLine import PipeLine
from libs.YOLO import YOLO26
from libs.Utils import *
import os,sys,gc
import ulab.numpy as np
import image

if __name__=="__main__":
    # This is only an example, please modify it to your own model path, label name, model input size, number of keypoints, and keypoint dimension for custom scenarios
    kmodel_path="/sdcard/examples/kmodel/yolo26n-pose.kmodel"
    labels = ["person"]
    model_input_size=[320,320]
    kp_num=17
    kp_dim=3

    # Display mode, "auto" automatically selects screen driver according to development board model; can also explicitly specify hdmi/lcd/st7701, etc., see PipeLine manual for values
    display_mode="auto"
    # None means using the default resolution of the screen driver; fill in [width, height] when specifying
    display_size=None
    rgb888p_size=[320,320]
    confidence_threshold = 0.5
    pl=PipeLine(rgb888p_size=rgb888p_size,display_mode=display_mode,display_size=display_size)
    pl.create()
    # Get the actual display resolution after create
    display_size=pl.get_display_size()
    # Initialize YOLO26 model
    yolo=YOLO26(task_type="pose",mode="video",kmodel_path=kmodel_path,labels=labels,rgb888p_size=rgb888p_size,model_input_size=model_input_size,display_size=display_size,kp_num=kp_num,kp_dim=kp_dim,conf_thresh=confidence_threshold,max_boxes_num=50,debug_mode=0)
    yolo.config_preprocess()
    while True:
        with ScopedTiming("total",1):
            img=pl.get_frame()
            res=yolo.run(img)
            yolo.draw_result(res,pl.osd_img)
            pl.show_image()
            gc.collect()
    yolo.deinit()
    pl.destroy()

The above code shows the code for using YOLO26 for video inference.