DUT<T>

Models & Types · AiDotNet.Video.Stabilization

DUT deep unsupervised trajectory-based video stabilization.

Inheritance: NeuralNetworkBase<T> → VideoNeuralNetworkBase<T> → VideoStabilizationBase<T> → DUT<T>

For Beginners

DUT (Deep Unsupervised Trajectory) stabilizes shaky video by learning smooth camera trajectories without requiring ground truth stabilized footage for training.

How It Works

References:

  • Paper: “DUT: Learning Video Stabilization by Simply Watching Unstable Videos” (Xu et al., ICCV 2022)

DUT learns to stabilize video in an unsupervised manner by predicting per-pixel flow fields from unstable videos alone. It uses a coarse-to-fine pyramid with temporal consistency loss to produce smooth warping fields without requiring paired stable/unstable training data.

Example

// Create a DUT model for unsupervised video stabilization
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.Generative,
    inputHeight: 256, inputWidth: 256, inputDepth: 3);
var options = new DUTOptions();
var dut = new DUT<double>(architecture, options);
            
// Or load a pre-trained ONNX model for inference
var dutOnnx = new DUT<double>(architecture, "dut_model.onnx");

Constructors

Constructor Summary
DUT(NeuralNetworkArchitecture<T> architecture, DUTOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) Creates a DUT model for native training and inference.
DUT(NeuralNetworkArchitecture<T> architecture, string modelPath, DUTOptions options) Creates a DUT model for native training and inference.

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> Stabilize(Tensor<T> unstableFrames)