VideoGigaGANOptions

Options & Configuration · AiDotNet.Video.Options

Configuration options for the VideoGigaGAN large-scale GAN for video SR.

Inheritance: ModelOptions → NeuralNetworkOptions → VideoGigaGANOptions

For Beginners

VideoGigaGAN is like having a very talented artist with a magnifying glass. While diffusion models (like StableVideoSR) gradually “develop” a high-res image from noise, VideoGigaGAN directly “paints” detail in a single forward pass – much faster. A special “detail shuttle” ensures real details are preserved while fake ones are avoided, and anti-aliasing prevents flickering.

How It Works

VideoGigaGAN (Xu et al., CVPR 2025) is the first large-scale GAN for video SR:

  • GigaGAN backbone: upscaled StyleGAN architecture with 1B+ parameters, providing

exceptional detail generation capability beyond diffusion models in sharpness

  • Feature propagation with anti-aliasing: temporal feature propagation using

anti-aliased flow warping to prevent temporal aliasing artifacts

  • High-frequency shuttle: a dedicated pathway that extracts and preserves high-frequency

details (edges, textures) from the input through a parallel processing stream, preventing the GAN from hallucinating false details

  • Temporal discriminator: a 3D discriminator that evaluates temporal consistency

in addition to per-frame quality

  • Supports up to 8x upscaling with rich perceptual details

Constructors

Constructor Summary
VideoGigaGANOptions() Initializes a new instance with default values.
VideoGigaGANOptions(VideoGigaGANOptions 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 GANWeight { get; set; } Gets or sets the weight for the GAN adversarial loss component.
double HFShuttleWeight { get; set; } Gets or sets the weight for the high-frequency shuttle loss.
double LearningRate { get; set; } Gets or sets the learning rate.
double PerceptualWeight { get; set; } Gets or sets the weight for the perceptual (LPIPS) loss component.
int FlowPyramidLevels { get; set; } Gets or sets the number of coarse-to-fine SPyNet flow-pyramid levels.
int NumFeatures { get; set; } Gets or sets the number of feature channels in the generator backbone.
int NumResBlocks { get; set; } Gets or sets the number of residual blocks in the generator.
int NumStyleLayers { get; set; } Gets or sets the number of style mixing layers in the GigaGAN generator.
int ScaleFactor { get; set; } Gets or sets the spatial upscaling factor.
string ModelPath { get; set; } Gets or sets the path to the ONNX model file.