FlowFormerPlusPlus<T>

Models & Types · AiDotNet.Video.Motion

FlowFormer++ masked cost volume autoencoding with tile-based high-resolution flow.

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

For Beginners

FlowFormer++ improves on FlowFormer with enhanced cost volume processing and better handling of occlusions. It achieves state-of-the-art accuracy on optical flow benchmarks.

How It Works

References:

  • Paper: “FlowFormer++: Masked Cost Volume Autoencoding for Pretraining Optical Flow Estimation” (Shi et al., 2023)

FlowFormer++ extends FlowFormer with masked cost volume autoencoding pretraining and tile-based processing for high-resolution optical flow.

Example

// Create a FlowFormer++ model for high-resolution optical flow
var flowFormerPP = new FlowFormerPlusPlus<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 FlowFormerPlusPlus<double>(architecture);

Constructors

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

Methods

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