AudioVisualCorrespondenceOptions
Options & Configuration · AiDotNet.NeuralNetworks.Options
Configuration options for the AudioVisualCorrespondenceNetwork.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → VisionLanguageInputOptions → VisionLanguageModelOptions → AudioVisualCorrespondenceOptions
How It Works
NumEncoderLayers specifies the actual number
of shared Dense/LayerNorm/Tanh encoder blocks. The default remains six; the former factory
ignored this value and always built four blocks. EmbeddingDimension
is the actual final encoder width. Correcting that topology changes the layer/parameter
layout of checkpoints made with the old factory; retraining or an explicit checkpoint
migration is required rather than treating the old four-block state as a six-block model.
Channels controls visual patch inputs and
defaults to three. It is validated when a visual modality API is used; feature-input
Predict/Train do not consume channel geometry. Frames must be channel-first and their spatial dimensions divisible by
sixteen. Custom architecture layers retain their Predict/Train behavior, but do not declare
the modality encoder/fusion roles required by the audio/visual embedding APIs.
Constructors
| Constructor | Summary |
|---|---|
AudioVisualCorrespondenceOptions() |
Initializes a new instance of the AudioVisualCorrespondenceOptions class carrying this model’s shipped defaults. |
AudioVisualCorrespondenceOptions(AudioVisualCorrespondenceOptions other) |
Copies the audio/video rates and every inherited configuration setting. |
Properties
| Property | Summary |
|---|---|
double VideoFrameRate { get; set; } |
Gets or sets the nominal rate associated with the supplied video frames. |
int AudioSampleRate { get; set; } |
Gets or sets the waveform sampling rate used to construct the model’s log-mel front end. |
Methods
| Method | Summary |
|---|---|
void Validate() |
Throws if a value this model requires has been left unset or is not positive. |