UNINEXT<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Foundation

UNINEXT: Universal Instance Perception as Object Discovery and Retrieval.

Inheritance: NeuralNetworkBase<T> → SegmentationModelBase<T> → PanopticSegmentationBase<T> → UNINEXT<T>

For Beginners

UNINEXT reformulates over 10 different instance perception tasks into a unified “discover and retrieve” framework. Whether you need object detection, instance segmentation, single-object tracking, multi-object tracking, video instance segmentation, or referring expression segmentation, UNINEXT handles them all with one model by using task-specific prompt embeddings.

Common use cases:

  • Multi-task instance perception (detection + segmentation + tracking)
  • Video object segmentation with tracking
  • Referring expression comprehension (find object from text description)
  • SOTA results on 20+ benchmarks simultaneously

How It Works

Technical Details:

  • Unified query representation for all instance perception tasks
  • Task-specific prompt embeddings select which task to perform
  • Backbone: ResNet-50, Swin-L, or ViT-H
  • Deformable transformer encoder-decoder
  • SOTA on 20+ benchmarks across detection, segmentation, tracking

Reference: Yan et al., “Universal Instance Perception as Object Discovery and Retrieval”, CVPR 2023.

Example

// Create a UNINEXT model for unified instance perception (detection + segmentation + tracking)
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.ThreeDimensional,
    taskType: NeuralNetworkTaskType.MultiClassClassification,
    inputHeight: 512, inputWidth: 512, inputDepth: 3, outputSize: 80);
var model = new UNINEXT<double>(architecture, options: new UNINEXTOptions { NumClasses = 80 });
            
// Or load a pre-trained ONNX model for multi-task instance perception
var onnxModel = new UNINEXT<double>(architecture, "uninext.onnx", options: new UNINEXTOptions { NumClasses = 80 });

Constructors

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

Methods

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