EdgeSAM<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Efficient

EdgeSAM: Prompt-in-the-Loop Distillation for SAM on edge devices.

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

For Beginners

Edge-device segment anything. On-device interactive segmentation.

Common use cases:

  • Edge-device segment anything
  • On-device interactive segmentation
  • Mobile and IoT segmentation
  • Real-time promptable segmentation

How It Works

Technical Details:

  • RepViT encoder for edge-optimized inference
  • Prompt-in-the-loop distillation from SAM ViT-H
  • End-to-end encoder + decoder distillation
  • Runs on iPhone 14 at interactive speeds

Reference: Zhou et al., “EdgeSAM: Prompt-In-the-Loop Distillation for On-Device Deployment of SAM”, arXiv 2024.

Example

// Create an EdgeSAM model for edge-device interactive segmentation
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.ImageSegmentation,
    inputHeight: 1024, inputWidth: 1024, inputDepth: 3, outputSize: 1);
var model = new EdgeSAM<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new EdgeSAM<double>(architecture,
    options: new EdgeSAMOptions { NumClasses = 1, DropRate = 0 });
            
// Or load a pre-trained ONNX model for mobile deployment
var onnxModel = new EdgeSAM<double>(architecture, "edgesam_model.onnx");

Constructors

Constructor Summary
EdgeSAM(NeuralNetworkArchitecture<T> architecture, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction, EdgeSAMOptions options) Initializes EdgeSAM in native (trainable) mode.
EdgeSAM(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, EdgeSAMOptions options) Initializes EdgeSAM 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)
void SetImage(Tensor<T> image)