EfficientSAM<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Efficient

EfficientSAM: Leveraged Masked Image Pretraining for efficient SAM.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → PromptableSegmentationBase<T> → EfficientSAM<T>

For Beginners

Efficient segment anything with strong pre-training. Interactive segmentation.

Common use cases:

  • Efficient segment anything with strong pre-training
  • Interactive segmentation
  • Automatic mask generation
  • Production SAM deployment

How It Works

Technical Details:

  • SAMI (SAM Image) pre-training with masked autoencoder
  • Lightweight ViT encoder with MAE knowledge transfer
  • Cross-attention decoder for mask prediction
  • Better quality-efficiency trade-off than MobileSAM

Reference: Xiong et al., “EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anything”, CVPR 2024.

Example

// Create an EfficientSAM model for efficient promptable segmentation
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.ImageSegmentation,
    inputHeight: 1024, inputWidth: 1024, inputDepth: 3, outputSize: 1);
var model = new EfficientSAM<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new EfficientSAM<double>(architecture,
    options: new EfficientSAMOptions { NumClasses = 1, DropRate = 0 });

Constructors

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

Methods

Method Summary
List<PromptedSegmentationResult<T>> SegmentEverything()
ModelMetadata<T> GetModelMetadata() Collects metadata describing this model’s configuration.
PromptedSegmentationResult<T> SegmentFromBox(Tensor<T> box)
PromptedSegmentationResult<T> SegmentFromMask(Tensor<T> mask)
PromptedSegmentationResult<T> SegmentFromPoints(Tensor<T> points, Tensor<T> labels)
Tensor<T> ForwardForTraining(Tensor<T> input) Uses the same encoder/decoder graph for autodiff training as inference.
void SetImage(Tensor<T> image)