VMamba<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Mamba

VMamba: Visual State Space Model with Cross-Scan for 2D understanding.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → SemanticSegmentationBase<T> → VMamba<T>

For Beginners

Hierarchical visual feature extraction. Semantic segmentation backbone.

Common use cases:

  • Hierarchical visual feature extraction
  • Semantic segmentation backbone
  • Dense prediction with global receptive field
  • Efficient alternative to Swin Transformer

How It Works

Technical Details:

  • Cross-Scan Module (CSM) for 2D-aware state space scanning
  • 4-way scanning to capture all spatial relationships
  • Hierarchical architecture similar to Swin Transformer
  • Linear complexity with global effective receptive field

Reference: Liu et al., “VMamba: Visual State Space Model”, NeurIPS 2024.

Example

// Create a VMamba model for hierarchical semantic segmentation
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.MultiClassClassification,
    inputHeight: 512, inputWidth: 512, inputDepth: 3, outputSize: 150);
var model = new VMamba<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new VMamba<double>(architecture,
    options: new VMambaOptions { NumClasses = 150, DropRate = 0.1, ModelSize = VMambaModelSize.Tiny });
            
// Or load a pre-trained ONNX model with cross-scan mechanism
var onnxModel = new VMamba<double>(architecture, "vmamba.onnx");

Constructors

Constructor Summary
VMamba(NeuralNetworkArchitecture<T> architecture, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction, VMambaOptions options) Initializes VMamba in native (trainable) mode.
VMamba(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, VMambaOptions options) Initializes VMamba in ONNX (inference-only) mode.

Methods

Method Summary
ModelMetadata<T> GetModelMetadata() Collects metadata describing this model’s configuration.