BSVDOptions

Options & Configuration · AiDotNet.Video.Options

Configuration options for BSVD bidirectional streaming video denoising.

Inheritance: ModelOptions → NeuralNetworkOptions → BSVDOptions

For Beginners

BSVD cleans up noisy video in real-time by looking at both past and future frames. Unlike methods that need all frames at once, it processes them in a stream using small memory buffers, making it practical for live video and long recordings.

How It Works

BSVD (Qi et al., ACM MM 2022) enables real-time video denoising through bidirectional streaming with efficient buffer management:

  • Bidirectional streaming: processes video in both forward and backward passes with shared

buffers, so each frame benefits from both past and future context

  • Streaming buffers: maintains compact latent buffers instead of storing full frames,

enabling constant-memory processing regardless of video length

  • Real-time capability: designed for 30+ fps denoising on consumer GPUs through

efficient buffer reuse and single-pass-per-direction processing

  • Noise-adaptive: handles varying noise levels without requiring noise-level input,

making it suitable for real-world video with spatially varying noise

Constructors

Constructor Summary
BSVDOptions() Initializes a new instance with default values.
BSVDOptions(BSVDOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
OnnxModelOptions OnnxOptions { get; set; } Gets or sets the ONNX runtime options.
VideoModelVariant Variant { get; set; } Gets or sets the model variant.
double AdamBeta1 { get; set; } Adam first-moment decay.
double AdamBeta2 { get; set; } Adam second-moment decay.
double DropoutRate { get; set; } Gets or sets the dropout rate.
double GradientClipNorm { get; set; } Maximum gradient norm applied during training.
double LearningRate { get; set; } Gets or sets the learning rate.
int BufferDim { get; set; } Gets or sets the hidden state dimension for streaming buffers.
int NumFeatures { get; set; } Gets or sets the number of feature channels.
int NumLevels { get; set; } Gets or sets the number of U-Net encoder/decoder levels.
int NumRecurrentBlocks { get; set; } Gets or sets the number of recurrent blocks per direction.
int NumUNetStages { get; set; } Gets or sets the number of sequential U-Nets in the denoising backbone.
int ShiftedChannelRatio { get; set; } Gets or sets the shifted-channel ratio r used by temporal fusion.
string ModelPath { get; set; } Gets or sets the path to the ONNX model file.