SEARAFT<T>

Models & Types · AiDotNet.Video.Motion

SEA-RAFT simple efficient accurate RAFT with mixture of Laplace loss.

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

For Beginners

SEA-RAFT extends RAFT with Scale-Equivariant Architecture for better handling of objects at different scales. It improves flow accuracy for both small and large motions.

How It Works

References:

  • Paper: “SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow” (Teed et al., ECCV 2024 Oral, Best Paper Candidate)

SEA-RAFT simplifies RAFT with a mixture of Laplace loss and direct initial flow prediction, achieving state-of-the-art accuracy with improved efficiency.

Example

// Create a SEA-RAFT model for simple, efficient optical flow
var seaRaft = new SEARAFT<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 SEARAFT<double>(architecture);

Constructors

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

Methods

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