Class NeckBase<T>

Namespace

AiDotNet.ComputerVision.Detection.Necks

Assembly

AiDotNet.dll

Base class for neck modules that perform multi-scale feature fusion.

public abstract class NeckBase<T>

Type Parameters

T

The numeric type used for calculations.

Inheritance

object

NeckBase<T>

Derived

BiFPN<T>

FPN<T>

PANet<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

For Beginners: The "neck" sits between the backbone and the detection head. It takes multi-scale features from the backbone and fuses them together so that each feature level contains information from both higher and lower resolutions. This helps detect objects of various sizes more accurately.

Common neck architectures: - FPN (Feature Pyramid Network): Top-down feature fusion - PANet (Path Aggregation Network): Top-down + bottom-up paths - BiFPN (Bidirectional FPN): Weighted bidirectional fusion

Constructors

NeckBase()

Creates a new neck module.

protected NeckBase()

Fields

IsTrainingMode

Whether the neck is in training mode.

protected bool IsTrainingMode

Field Value

bool

NumOps

Numeric operations for type T.

protected readonly INumericOperations<T> NumOps

Field Value

INumericOperations<T>

Properties

Name

Name of this neck architecture.

public abstract string Name { get; }

Property Value

string

NumLevels

Number of feature levels output by the neck.

public abstract int NumLevels { get; }

Property Value

int

OutputChannels

Number of output channels for all feature levels.

public abstract int OutputChannels { get; }

Property Value

int

Remarks

Necks typically project all feature levels to the same number of channels (e.g., 256) to simplify the detection head.

Methods

Add(Tensor<T>, Tensor<T>)

Adds two feature maps element-wise.

protected Tensor<T> Add(Tensor<T> a, Tensor<T> b)

Parameters

a Tensor<T>

First feature map.

b Tensor<T>

Second feature map.

Returns

Tensor<T>

Element-wise sum.

Conv1x1(Tensor<T>, Tensor<T>, Tensor<T>?)

Applies a 1x1 convolution to change the number of channels.

protected Tensor<T> Conv1x1(Tensor<T> input, Tensor<T> weights, Tensor<T>? bias = null)

Parameters

input Tensor<T>

Input feature map.

weights Tensor<T>

Convolution weights [out_channels, in_channels].

bias Tensor<T>

Optional bias [out_channels].

Returns

Tensor<T>

Feature map with new channel count.

Downsample2x(Tensor<T>)

Downsample a feature map by a factor of 2 using max pooling.

protected Tensor<T> Downsample2x(Tensor<T> input)

Parameters

input Tensor<T>

Input feature map.

Returns

Tensor<T>

Downsampled feature map.

Forward(List<Tensor<T>>)

Performs multi-scale feature fusion.

public abstract List<Tensor<T>> Forward(List<Tensor<T>> features)

Parameters

features List<Tensor<T>>

List of feature maps from the backbone, ordered from highest to lowest resolution.

Returns

List<Tensor<T>>

Fused feature maps at multiple scales.

Remarks

For Beginners: This method takes the raw features from the backbone and combines them across scales. After fusion, each feature level "knows about" features from other scales, making detection more accurate.

GetParameterCount()

Gets the total number of parameters in the neck.

public abstract long GetParameterCount()

Returns

long

Number of trainable parameters.

ReadParameters(BinaryReader)

Reads parameters from a binary reader for deserialization.

public abstract void ReadParameters(BinaryReader reader)

Parameters

reader BinaryReader

The binary reader to read from.

SetTrainingMode(bool)

Sets whether the neck is in training mode.

public virtual void SetTrainingMode(bool training)

Parameters

training bool

True for training, false for inference.

Upsample2x(Tensor<T>)

Upsample a feature map by a factor of 2 using nearest neighbor interpolation.

protected Tensor<T> Upsample2x(Tensor<T> input)

Parameters

input Tensor<T>

Input feature map.

Returns

Tensor<T>

Upsampled feature map.

ValidateFeatures(List<Tensor<T>>, int[])

Validates that the input features are compatible with this neck.

protected void ValidateFeatures(List<Tensor<T>> features, int[] expectedInputChannels)

Parameters

features List<Tensor<T>>

Features to validate.

expectedInputChannels int[]

Expected input channels at each level.

Exceptions

ArgumentException

Thrown if features are incompatible.

WriteParameters(BinaryWriter)

Writes all parameters to a binary writer for serialization.

public abstract void WriteParameters(BinaryWriter writer)

Parameters

writer BinaryWriter

The binary writer to write to.