DPFlow<T>

Models & Types · AiDotNet.Video.Motion

DPFlow dual-pyramid framework combining image and feature pyramid advantages.

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

For Beginners

DPFlow (Dual-Path Flow) estimates optical flow using parallel spatial and temporal processing paths. The dual-path design captures both fine spatial detail and broad temporal motion patterns.

How It Works

References:

  • Paper: “DPFlow: Dual-Pyramid Optical Flow” (Morimitsu et al., CVPR 2025)

DPFlow combines the advantages of image pyramids (capturing large motions) and feature pyramids (rich semantics) in a unified dual-pyramid framework.

Example

// Create a DPFlow model for dual-pyramid optical flow estimation
var dpFlow = new DPFlow<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 DPFlow<double>(architecture);

Constructors

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

Methods

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