YOLOv12Seg<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.InstanceSegmentation

YOLOv12-Seg: Attention-centric real-time instance segmentation.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → InstanceSegmentationBase<T> → YOLOv12Seg<T>

For Beginners

Real-time instance segmentation with attention. Autonomous vehicle perception.

Common use cases:

  • Real-time instance segmentation with attention
  • Autonomous vehicle perception
  • Smart surveillance systems
  • Industrial quality control

How It Works

Technical Details:

  • Area-attention mechanism replacing standard self-attention
  • R-ELAN (Residual Efficient Layer Aggregation Network) blocks
  • FlashAttention-compatible efficient attention
  • Native attention integration in YOLO architecture

Reference: Tian et al., “YOLOv12: Attention-Centric Real-Time Object Detectors”, arXiv 2025.

Example

// Create a YOLOv12-Seg model with attention-centric instance segmentation
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.MultiClassClassification,
    inputHeight: 640, inputWidth: 640, inputDepth: 3, outputSize: 80);
var model = new YOLOv12Seg<double>(architecture);
// Every tunable is now set through the options object; these are the defaults:
var custom = new YOLOv12Seg<double>(architecture,
    options: new YOLOv12SegOptions { NumClasses = 80, DropRate = 0, ModelSize = YOLOv12SegModelSize.N });
            
// Or load a pre-trained ONNX model for autonomous vehicle perception
var onnxModel = new YOLOv12Seg<double>(architecture, "yolov12n-seg.onnx");

Constructors

Constructor Summary
YOLOv12Seg(NeuralNetworkArchitecture<T> architecture, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction, YOLOv12SegOptions options) Initializes YOLOv12Seg in native (trainable) mode.
YOLOv12Seg(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, YOLOv12SegOptions options) Initializes YOLOv12Seg in ONNX (inference-only) mode.

Methods

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