TemporalConv3DLayer<T>
Layers ยท AiDotNet.NeuralNetworks.Layers
A trainable anisotropic 3D convolution for video tensors in NCFHW layout.
Inheritance: LayerBase<T> โ TemporalConv3DLayer<T>
Implements: IShapeContract
How It Works
Unlike Conv3DLayer, this layer permits different temporal and
spatial kernel sizes. It is intended for paper architectures whose temporal
kernels are (K,1,1). Forward execution is composed from
Conv3D, so
CPU execution, all direct-GPU backends, and gradient-tape registration share
the same implementation.
Constructors
| Constructor | Summary |
|---|---|
TemporalConv3DLayer(int inputChannels, int outputChannels, int kernelDepth, int kernelHeight, int kernelWidth, int? paddingDepth, int? paddingHeight, int? paddingWidth, bool zeroInitialize) |
Creates an anisotropic video convolution. |
Properties
| Property | Summary |
|---|---|
bool IsZeroInitialized { get; } |
Gets whether the layer is initialized as an exact zero projection. |
bool SupportsTraining { get; } |
|
int InputChannels { get; } |
Gets the input channel count. |
int KernelDepth { get; } |
Gets the temporal kernel extent. |
int KernelHeight { get; } |
Gets the spatial kernel height. |
int KernelWidth { get; } |
Gets the spatial kernel width. |
int OutputChannels { get; } |
Gets the output channel count. |
Methods
| Method | Summary |
|---|---|
Dictionary<string, string> GetParameterRoles() |
Returns parameter roles for per-role optimizer configuration (e.g., weight decay exemption for biases). |
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
|
IReadOnlyList<Tensor<T>> GetTrainableParameters() |
Returns all trainable parameter tensors marked with [TrainableParameter]. |
void ResetState() |
|
void SetTrainableParameters(IReadOnlyList<Tensor<T>> parameters) |
Replaces trainable parameter tensors (e.g., with ParameterBuffer views). |
void UpdateParameters(T learningRate) |
|
void ZeroGrad() |
Clears all gradient fields discovered by convention ({paramName}Gradient). |