RPKNet<T>

Models & Types · AiDotNet.Video.Motion

RPKNet recurrent partial kernel network with separable large kernels for flow.

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

For Beginners

RPKNet (Recurrent Position-aware Kernel Network) uses position-aware convolution kernels that adapt to each pixel position for accurate optical flow estimation.

How It Works

References:

  • Paper: “RPKNet: Recurrent Partial Kernel Network for Efficient Optical Flow” (Morimitsu et al., AAAI 2024)

RPKNet uses recurrent partial kernel processing with separable large kernels for variable multi-scale feature extraction in optical flow.

Example

// Create an RPKNet model for partial kernel optical flow estimation
var rpkNet = new RPKNet<double>();
            
// Or configure with custom parameters
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.Regression,
    inputHeight: 256, inputWidth: 256, inputDepth: 3, outputSize: 2);
var model = new RPKNet<double>(architecture);

Constructors

Constructor Summary
RPKNet() Initializes a new instance with default architecture settings.
RPKNet(NeuralNetworkArchitecture<T> architecture, RPKNetOptions options) Creates a new RPKNet model for native training and inference.

Methods

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