VideoSuperResolutionBase<T>
Base Classes · AiDotNet.Video
Base class for video super-resolution models that upscale low-resolution video to higher resolution.
Inheritance: NeuralNetworkBase<T> → VideoNeuralNetworkBase<T> → VideoSuperResolutionBase<T>
Implements: IShapeContract, IVideoSuperResolution<T>
Known derived types: BasicVSRPlusPlus<T>, BasicVSR<T>, DAMVSR<T>, DOVE<T>, DualXVSR<T>, EDVR<T>, FlashVSR<T>, IART<T>, IconVSR<T>, MGLDVSR<T>, MIAVSR<T>, PSRT<T>, RVRT<T>, RealBasicVSRSharp<T>, RealBasicVSR<T>, RealESRGANVideo<T>, RealESRGAN<T>, RealViformer<T>, RealisVSR<T>, SeedVR<T>, StableVideoSR<T>, StreamDiffVSR<T>, TTVSR<T>, Upscale4KAgent<T>, VRT<T>, VideoGigaGAN<T>
For Beginners
Video super-resolution makes low-resolution video sharper and more detailed. For example, it can upscale a 480p video to 4K quality. Unlike single-image methods, video SR uses information from neighboring frames for better quality and temporal consistency (no flickering between frames).
How It Works
Video super-resolution extends image super-resolution by exploiting temporal information across multiple frames. This base class provides:
- Scale factor management (2x, 4x, 8x upscaling)
- Tile-based inference for memory-efficient processing of high-resolution video
- Bicubic upsampling as fallback/initialization
- Temporal consistency utilities
Derived classes implement specific architectures like BasicVSR++, RVRT, RealBasicVSR, etc.
Properties
| Property | Summary |
|---|---|
int ScaleFactor { get; set; } |
Gets the spatial upscaling factor (e.g., 2 for 2x, 4 for 4x). |
int TileOverlap { get; set; } |
Gets or sets the overlap between adjacent tiles. |
int TileSize { get; set; } |
Gets or sets the tile size for memory-efficient tiled processing. |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
The super-resolution family’s law: both spatial axes scale by ScaleFactor and every other axis is carried through unchanged. |
Tensor<T> EstimateFlow(Tensor<T> frame1, Tensor<T> frame2) |
Estimates optical flow between two frames for temporal alignment. |
Tensor<T> Upscale(Tensor<T> lowResFrames) |
Upscales a sequence of video frames. |