Object Detection(Recognize Orientation)

Function: Uses deep learning methods to segment and detect input images, usually processing box-like objects.

Input Parameters:

Name	Type	Valid Range	Default Value	Meaning
Image	ColorImage	None	None	Input image, requires a color image with RGB channels.

Name

Type

Valid Range

Default Value

Meaning

Image

ColorImage

None

Input image, requires a color image with RGB channels.

Output Parameters:

Name	Type	Valid Range	Default Value	Meaning
Detection results	DetectInstance	None	{}	Returns bounding boxes, categories, scores and polygon.

Name

Type

Valid Range

Default Value

Meaning

Detection results

DetectInstance

None

{}

Returns bounding boxes, categories, scores and polygon.

Parameter Settings:

Name	Type	Valid Range	Default Value	Meaning
Weight files	File	['.pth']	None	Model file.
Confidence threshold	Float	[0.0, 1.0]	0.5	Detection threshold.
Enable GPU	Bool	None	False	Set whether to use the GPU for inference, if it is turned on, the computer needs to have a graphics card environment.
Enable rotation	Bool	[True, False]	False	Set whether the model is trained to rotate. If the angle is trained, the resulting detection box will rotate to close to the target.
Legacy callout model	Bool	None	False	Set whether the model is a model trained on the old version of the annotation data. If true, it is a strict order four-point annotation. If false, it is a rotating rectangle annotation.

Name

Type

Valid Range

Default Value

Meaning

Weight files

File

['.pth']

None

Model file.

Confidence threshold

Float

[0.0, 1.0]

0.5

Detection threshold.

Enable GPU

Bool

None

False

Set whether to use the GPU for inference, if it is turned on, the computer needs to have a graphics card environment.

Enable rotation

Bool

[True, False]

False

Set whether the model is trained to rotate. If the angle is trained, the resulting detection box will rotate to close to the target.

Legacy callout model

Bool

None

False

Set whether the model is a model trained on the old version of the annotation data. If true, it is a strict order four-point annotation. If false, it is a rotating rectangle annotation.