DKM<T>

Models & Types · AiDotNet.Video.Motion

DKM dense kernelized feature matching for geometry estimation.

Inheritance: NeuralNetworkBase<T> → VideoNeuralNetworkBase<T> → OpticalFlowBase<T> → DKM<T>

For Beginners

DKM (Dense Kernelized Matching) estimates dense correspondences between image pairs using learned kernels. It produces accurate pixel-level matching for optical flow and stereo vision.

How It Works

References:

  • Paper: “DKM: Dense Kernelized Feature Matching for Geometry Estimation” (Edstedt et al., CVPR 2023)

DKM uses dense kernelized matching to establish pixel-level correspondences between images for accurate geometry estimation.

Example

// Create a DKM model for dense feature matching and geometry estimation
var dkm = new DKM<double>();
            
// Or configure with custom feature dimensions
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.Regression,
    inputHeight: 256, inputWidth: 256, inputDepth: 3, outputSize: 2);
var model = new DKM<double>(architecture, options: new DKMOptions { NumFeatures = 64, NumLayers = 8 });

Constructors

Constructor Summary
DKM() Creates a new DKM model for native training and inference.
DKM(NeuralNetworkArchitecture<T> architecture, DKMOptions options)

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> EstimateFlow(Tensor<T> frame0, Tensor<T> frame1)