SOLOv2<T>
Models & Types · AiDotNet.ComputerVision.Segmentation.InstanceSegmentation
SOLOv2 (Segmenting Objects by Locations v2) for instance segmentation.
Inheritance: ModelBase<T, Tensor<T>, Tensor<T>> → InstanceSegmenterBase<T> → SOLOv2<T>
Implements: IGeneratedParameterRegistrar<T>
For Beginners
SOLOv2 is an anchor-free, box-free instance segmentation method. It directly predicts instance masks by location, dividing the image into a grid and predicting masks for each cell. This eliminates the need for ROI operations.
How It Works
Key features:
- Direct mask prediction without boxes
- Dynamic convolution for mask generation
- Grid-based location encoding
- More efficient than two-stage methods
Reference: Wang et al., “SOLOv2: Dynamic and Fast Instance Segmentation”, NeurIPS 2020
Constructors
| Constructor | Summary |
|---|---|
SOLOv2(InstanceSegmentationOptions<T> options) |
Creates a new SOLOv2 model. |
Properties
| Property | Summary |
|---|---|
string Name { get; } |
Methods
| Method | Summary |
|---|---|
InstanceSegmentationResult<T> Segment(Tensor<T> image) |
|
Task LoadWeightsAsync(string pathOrUrl, CancellationToken cancellationToken) |
|
long GetParameterCount() |
|
void SaveWeights(string path) |