Object Detection

Query image and prediction

Currently, there are two YOLOv5 implementations avaialables for object detection task. Old implementation will be removed in future version.

Those models contains a large number of pretrained weights. Please refer to the following table for available options.
The first letter nano, s, ... is the scaling and suffix simp means the model is simplified by onnx simplifier. With the number 6, the model is updated to the upstream version of YOLOv5.

Model	Weights	Model	Weights
YOLO2	yolo_nano yolo_nano6 yolo_nano_simp yolo_nano6_simp yolo_s yolo_s6 yolo_s_simp yolo_s6_simp yolo_m yolo_m6 yolo_m_simp yolo_m6_simp yolo_l yolo_l_simp yolo_x yolo_x6 yolo_x_simp yolo_x6_simp	YOLO	yolo_nano yolo_s yolo_xl yolo_extreme yolo_nano_smp yolo_s_smp yolo_xl_smp yolo_extreme_smp

---

BiWAKO.YOLO2

Note

It is recommended to use this model rather than previous YOLO. This model optimizes pre/post-processing operations with new ort opsets. Runtime is 3~4 times faster than the previous model. If you want to use the raw output of the YOLO or customize post-processing with your choice of parameters, use the previous model below.

Faster implementation of YOLO without any pre/post-processing.

Attributes:

Name	Type	Description
model_path	str	The path to the onnx file. If automatic download is triggered, the file is downloaded to this path.
session	onnxruntime.InferenceSession	The session of the onnx model.
input_name	str	The name of the input node.
coco_label	List[str]	The coco mapping of the labels.
colors	Colors	Color palette written by Ultralytics at https://github.com/ultralytics/yolov5/blob/a3d5f1d3e36d8e023806da0f0c744eef02591c9b/utils/plots.py

__init__(self, model='yolo_nano_simp') special

Initialize YOLO2.

Parameters:

Name	Type	Description	Default
model	str	Choice of the model. Also accept the path to the downloaded onnx file. If the model has not been downloaded yet, the file is downloaded automatically. Defaults to "yolo_nano_simp".	'yolo_nano_simp'

Examples:

>>> model = YOLO2("yolo_nano_simp")

predict(self, image)

Return the prediction of the model.

Parameters:

Name	Type	Description	Default
image	Image	The image to be predicted. Accept both path and array in cv2 format.	required

Returns:

Type	Description
List[np.ndarray]	The prediction of the model in the format of [scores, labels, boxes].

Examples:

>>> model.predict("path/to/image.jpg")

render(self, prediction, image, **kwargs)

Return the original image with the predicted bounding boxes.

Parameters:

Name	Type	Description	Default
prediction	List[np.ndarray]	The prediction of the model in the format of [scores, labels, boxes].	required
image	Image	The image to be predicted. Accept both path and array in cv2 format.	required

Returns:

Type	Description
np.ndarray	The image with the predicted bounding boxes in cv2 format.

Examples:

>>> model.render(model.predict("path/to/image.jpg"), "path/to/image.jpg")

---

BiWAKO.YOLO

YOLOv5 onnx model.

Attributes:

Name	Type	Description
model_path	str	Path to the onnx file. If auto download is triggered, the file is downloaded to this path.
session	onnxruntime.InferenceSession	Inference session.
input_name	str	Name of the input node.
output_name	str	Name of the output node.
input_shape	tuple	Shape of the input image. Set accordingly to the model.
coco_label	list	List of coco 80 labels.
colors	Colors	Color palette written by Ultralytics at https://github.com/ultralytics/yolov5/blob/a3d5f1d3e36d8e023806da0f0c744eef02591c9b/utils/plots.py

__init__(self, model='yolo_nano') special

Initialize the model.

Parameters:

Name	Type	Description	Default
model	str	Model type to be used. Also accept path to the onnx file. If the model is not found, it will be downloaded automatically. Currently [yolo_nano, yolo_s, yolo_xl and yolo_extreme] are supported. Default is yolo_nano. Adding _smp to the model name will use the simplified model.	'yolo_nano'

Examples:

>>> model = YOLO("yolo_nano_smp")
downloading yolo_nano_smp.onnx to yolo_nano_smp.onnx
100%|██████████| 7.57M/7.57M [00:01<00:00, 6.89MB/s]

predict(self, image)

Return the prediction of the model.

Parameters:

Name	Type	Description	Default
image	Image	Image to be predicted. Accept str or cv2 image.	required

Returns:

Type	Description
np.ndarray	n by 6 array where 2nd dimension is xyxy with label and confidence.

render(self, prediction, image)

Return the original image with predicted bounding boxes.

Parameters:

Name	Type	Description	Default
prediction	np.ndarray	Prediction of the model.	required
image	Image	Image to be predicted. Accept str or cv2 image.	required

Returns:

Type	Description
np.ndarray	Image with predicted bounding boxes in cv2 format.