CATSeg<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.OpenVocabulary

CAT-Seg: Cost Aggregation for open-vocabulary semantic segmentation.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → OpenVocabSegmentationBase<T> → CATSeg<T>

For Beginners

Open-vocabulary segmentation via cost aggregation. Novel category recognition.

Common use cases:

  • Open-vocabulary segmentation via cost aggregation
  • Novel category recognition
  • Cross-dataset segmentation
  • Language-driven pixel classification

How It Works

Technical Details:

  • Cost volume between CLIP image and text features
  • Cost aggregation transformer for spatial refinement
  • Exploits CLIP feature similarity for pixel classification
  • Efficient inference with frozen CLIP backbone

Reference: Cho et al., “CAT-Seg: Cost Aggregation for Open-Vocabulary Semantic Segmentation”, CVPR 2024.

Example

// Create a CAT-Seg model for open-vocabulary semantic segmentation
// Any CLIP-style ITextEncoder supplies the text embeddings, e.g. a CLIP model from VisionLanguage.Encoders.
ITextEncoder<double> textEncoder = clipModel;
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.MultiClassClassification,
    inputHeight: 512, inputWidth: 512, inputDepth: 3, outputSize: 150);
var model = new CATSeg<double>(architecture, textEncoder);
// Every tunable is now set through the options object; these are the defaults:
var custom = new CATSeg<double>(architecture, textEncoder,
    options: new CATSegOptions { NumClasses = 150, DropRate = 0.1 });
            
// Or load a pre-trained ONNX model for novel category recognition
var onnxModel = new CATSeg<double>(architecture, "catseg.onnx", textEncoder, options: null);

Constructors

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

Methods

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