TemporalModule3DLayer<T>
Layers · AiDotNet.Diffusion.NoisePredictors
Released Upscale-A-Video temporal module: temporal 3D ResNet, inflated spatial ResNet, and a zero-initialized residual projection.
Inheritance: LayerBase<T> → TemporalModule3DLayer<T>
Implements: IShapeContract
How It Works
The released configuration disables both optional transformer-attention branches. Consequently this layer contains the exact convolutional path and remains independent of the number of frames supplied at runtime.
Constructors
| Constructor | Summary |
|---|---|
TemporalModule3DLayer(int channels, int timeEmbeddingDim, int spatialSize) |
Creates the paper/released-code temporal module. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
|
bool UsesTransformerAttention { get; } |
Gets whether optional transformer attention is present. |
bool UsesZeroInitializedOutputProjection { get; } |
Gets whether the residual output projection starts at zero. |
int FirstTemporalKernelDepth { get; } |
Gets the released first temporal kernel depth. |
int SecondTemporalKernelDepth { get; } |
Gets the released second temporal kernel depth. |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<ILayer<T>> GetSubLayers() |
|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
Derived from this type’s [TensorLayout] declarations. |
Tensor<T> Forward(Tensor<T> input, Tensor<T> timeEmbedding) |
Runs the temporal module with a projected diffusion timestep embedding. |
void ResetState() |
|
void SetTrainingMode(bool isTraining) |
|
void UpdateParameters(T learningRate) |