Neural

DeepNet

Main neural-network interface for training, detection, tracking, prediction, and export.

43 config keys · 16 public methods

Import

from Xenonix.Neural import DeepNet

Search this API

Filter methods, arguments, and configuration keys on this page.

Configuration

Use Config(...) with these keys where the class exposes configuration.

Config key	Default	Common values	What it controls
`Verbose`	`True`	`True`, `False`	Prints/logs progress and status messages.
`Model`	`'Medium'`	`Small`, `Medium`, `Large`, `XSmall`, `XLarge`, `custom path`	Model size/name/path selector.
`Variant`	`'B'`	`A`, `B`, `C`, `custom`	Model variant.
`Channels`	`128`	Any valid value	Neural feature channel count.
`Inference`	`'HBB'`	`classify`, `hbb`, `hbbv2`, `segment`, `pose`, `obb`, `obbv2`	Neural task type such as classify, detect, segment, pose, OBB, or HBB.
`GPU`	`True`	`True`, `False`	Uses GPU acceleration where available.
`Layer_Start`	`None`	Any valid value	First layer used for training/inference where supported.
`Layers`	`None`	Any valid value	Layer list/selection.
`FPN_Layers`	`None`	Any valid value	Feature-pyramid layers.
`Data`	`False`	Any valid value	Dataset path or dataset configuration.
`Val_Freq`	`1`	Any valid value	Validation frequency.
`Val_Subset`	`0`	Any valid value	Validation subset size/count.
`Image_Size`	`False`	Any valid value	Input image size.
`Learning_Rate`	`None`	Any valid value	Training learning rate.
`Epoch`	`20`	Any valid value	Number of training epochs.
`Batch_Size`	`2`	Any valid value	Training/inference batch size.
`Worker`	`4`	Any valid value	Number of data-loader workers.
`Patience`	`None`	Any valid value	Early-stop patience.
`Step`	`None`	Any valid value	Learning-rate step schedule.
`Gamma`	`0.1`	Any valid value	Learning-rate decay factor.
`Momentum`	`0.9`	Any valid value	Optimizer momentum.
`Weight_Decay`	`None`	Any valid value	Weight decay regularization.
`Callback`	`False`	Any valid value	Optional progress callback.
`Threshold`	`0.5`	Any valid value	Confidence threshold.
`NMS`	`0.5`	Any valid value	Non-maximum suppression overlap threshold.
`NMS_Batch`	`False`	Any valid value	Batch NMS setting.
`Test_Threshold`	`0.3`	Any valid value	Threshold used during testing.
`Margin`	`10`	Any valid value	Extra region margin.
`Max_Age`	`30`	Any valid value	Maximum track age before dropping a track.
`Pretrained`	`True`	`True`, `False`	Uses pretrained weights where available.
`Export_Opset`	`20`	Any valid value	ONNX export opset version.
`Mask_Quality`	`False`	Any valid value	Mask quality setting.
`Mask_Layers`	`4`	Any valid value	Number of mask layers.
`Detections`	`300`	Any valid value	Maximum detections to keep.
`TopK`	`1000`	Any valid value	Top-K candidates to keep.
`Pre_Top`	`2000`	Any valid value	Candidates kept before filtering.
`Post_Top`	`1000`	Any valid value	Candidates kept after filtering.
`Text_Region`	`False`	Any valid value	Draws/shows region text.
`Text_Class`	`True`	Any valid value	Draws/shows class text.
`Text_Score`	`False`	Any valid value	Draws/shows score text.
`Text_ID`	`False`	Any valid value	Draws/shows track ID text.
`Anchor_Scale`	`None`	Any valid value	Anchor scale list/setting.
`Anchor_Ratio`	`None`	Any valid value	Anchor ratio list/setting.

Public methods

Create

Creates a new DeepNet instance.

Create(Path=False)

Arguments

Argument	Default	Accepted values	What it does
`Path`	`False`	Any valid value	File path used for reading or writing.

Output

Creates a new DeepNet object.

Config_Get

Reads one or more configuration values by name.

Config_Get(*Input)

Arguments

Argument	Default	Accepted values	What it does
`Input`	`*args`	Any valid value	Input value used by the operation.

Output

Returns selected configuration values.

Config

Changes one or more supported settings. Pass only the configuration keys you want to update.

Config(**Input)

Arguments

Argument	Default	Accepted values	What it does
`Input`	`**kwargs`	Any valid value	Input value used by the operation.

Output

Updates configuration values on the object.

Pause

Requests the active training or long-running neural process to pause.

Pause()

Arguments

No arguments.

Output

Runs the operation. Some methods update the object in place.

Resume

Resumes a paused neural process.

Resume()

Arguments

No arguments.

Output

Runs the operation. Some methods update the object in place.

Exit

Requests the active neural process to stop/exit cleanly.

Exit()

Arguments

No arguments.

Output

Runs the operation. Some methods update the object in place.

Test

Runs the current trainer test/validation routine.

Test()

Arguments

No arguments.

Output

Returns the calculated result.

Matrix

Builds or shows the confusion matrix for the current trainer.

Matrix()

Arguments

No arguments.

Output

Runs the operation. Some methods update the object in place.

Train

Starts training using the configured model, dataset, and inference type.

Train()

Arguments

No arguments.

Output

Runs the operation. Some methods update the object in place.

Detect

Runs object detection on a frame and returns detections/overlay data.

Detect(Frame)

Arguments

Argument	Default	Accepted values	What it does
`Frame`	Required	Any valid value	Input image/frame data. Can be an image object, array, or frame accepted by this tool.

Output

Returns the calculated result.

Track

Runs detection with tracking state and returns tracked results.

Track(Frame)

Arguments

Argument	Default	Accepted values	What it does
`Frame`	Required	Any valid value	Input image/frame data. Can be an image object, array, or frame accepted by this tool.

Output

Returns the calculated result.

Predict

Runs the model prediction pipeline on a frame.

Predict(Frame)

Arguments

Argument	Default	Accepted values	What it does
`Frame`	Required	Any valid value	Input image/frame data. Can be an image object, array, or frame accepted by this tool.

Output

Returns the calculated result.

Export

Exports the configured/trained model, typically to ONNX.

Export()

Arguments

No arguments.

Output

Returns the calculated result.

Accuracy

Returns or displays the accuracy metric helper.

Accuracy()

Arguments

No arguments.

Output

Returns the calculated result.

Performance

Displays performance metric helper/output.

Performance(Fixed=True)

Arguments

Argument	Default	Accepted values	What it does
`Fixed`	`True`	Any valid value	Turns fixed on or off.

Output

Runs the operation. Some methods update the object in place.

Precision

Displays precision metric helper/output.

Precision(Fixed=True)

Arguments

Argument	Default	Accepted values	What it does
`Fixed`	`True`	Any valid value	Turns fixed on or off.

Output

Runs the operation. Some methods update the object in place.