FastSAM<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Efficient

FastSAM: Fast Segment Anything Model based on YOLOv8-Seg.

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

For Beginners

Real-time segment anything. Interactive segmentation on edge devices.

Common use cases:

  • Real-time segment anything
  • Interactive segmentation on edge devices
  • Fast automatic mask generation
  • Lightweight alternative to SAM

How It Works

Technical Details:

  • YOLOv8-Seg backbone trained on SA-1B dataset
  • 50x faster than original SAM
  • CNN-based architecture (no vision transformer)
  • Supports point and box prompts

Reference: Zhao et al., “Fast Segment Anything”, arXiv 2023.

Example

// Create a FastSAM model for real-time segment anything via YOLOv8
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.ImageSegmentation,
    inputHeight: 1024, inputWidth: 1024, inputDepth: 3, outputSize: 1);
var model = new FastSAM<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new FastSAM<double>(architecture,
    options: new FastSAMOptions { NumClasses = 1, DropRate = 0 });

Constructors

Constructor Summary
FastSAM(NeuralNetworkArchitecture<T> architecture, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction, FastSAMOptions options) Initializes FastSAM in native (trainable) mode.
FastSAM(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, FastSAMOptions options) Initializes FastSAM 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 native encoder/decoder graph for tape-based training.
void SetImage(Tensor<T> image)