LiteDVDNetOptions

Options & Configuration · AiDotNet.Video.Options

Configuration options for LiteDVDNet lightweight deep video denoising.

Inheritance: ModelOptions → NeuralNetworkOptions → LiteDVDNetOptions

For Beginners

LiteDVDNet is a fast, lightweight video denoiser. It first cleans each frame individually, then combines information from nearby frames to improve quality. It’s designed to run efficiently on devices with limited computing power.

How It Works

LiteDVDNet (Ilchenko & Stirenko, IJIGSP 2025) optimizes FastDVDnet for speed via four changes:

  • Caching intermediate denoising results across overlapping frame windows (INFERENCE only - the paper

notes that mode “is suitable only for computations, not for network training”)

  • Reducing the InputCvBlock’s intermediate channels by a factor of three (90 -> 30)
  • Simplifying each convolutional block to a single convolution (“LiteCvBlock”)
  • Halving the channel count for the smaller variant (2.48M -> 0.64M parameters)

It keeps FastDVDnet’s Conv -> BatchNorm -> ReLU ordering, its PixelShuffle upsampling, and its residual connection between the central noisy input frame and the output. NOTE: earlier documentation here described depthwise separable convolutions and 1x1 temporal fusion. Neither appears anywhere in the paper.

Constructors

Constructor Summary
LiteDVDNetOptions() Initializes a new instance with default values.
LiteDVDNetOptions(LiteDVDNetOptions 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 DropoutRate { get; set; } Gets or sets the dropout rate.
double LearningRate { get; set; } Gets or sets the learning rate.
double ResidualHeadInitScale { get; set; } Gets or sets the scale applied to the residual-prediction head’s initial weights.
int ExpansionFactor { get; set; } Gets or sets a legacy expansion factor.
int InputBlockIntermediateChannels { get; set; } Gets or sets the InputCvBlock’s intermediate channel count.
int NumBlocks { get; set; } Gets or sets the number of denoising blocks per stage.
int NumFeatures { get; set; } Gets or sets the base feature width, which is the paper’s variant number: 32 for LiteDVDNet-32 (the balanced variant, 3x faster than FastDVDnet for -0.18 dB PSNR) and 16 for LiteDVDNet-16 (5x faster, -0.61 dB).
int TemporalWindowSize { get; set; } Gets or sets the temporal window size (number of input frames).
string ModelPath { get; set; } Gets or sets the path to the ONNX model file.