PointCloudSegmentationOutput<T>

Models & Types · AiDotNet.ComputerVision.Segmentation.Common

Output for 3D point cloud segmentation.

For Beginners

Point cloud segmentation labels each 3D point (from LiDAR, depth cameras, etc.) with a class. This output contains per-point labels, confidences, and optional per-point features for downstream tasks like scene understanding.

Constructors

Constructor Summary
PointCloudSegmentationOutput()

Properties

Property Summary
Dictionary<int, int> ClassPointCounts { get; set; } Per-class point counts.
Dictionary<int, string> ClassNames { get; set; } Per-class names (if available).
Tensor<T> Confidences { get; set; } Per-point confidence scores [N].
Tensor<T> Features { get; set; } Per-point learned feature vectors [N, featureDim] from the model’s encoder.
Tensor<T> InstanceBoundingBoxes { get; set; } Bounding boxes for each detected instance [numInstances, 6] as (x1, y1, z1, x2, y2, z2).
Tensor<T> InstanceIds { get; set; } Per-point instance IDs [N] for instance segmentation (if supported).
Tensor<T> Labels { get; set; } Per-point class labels [N].
Tensor<T> Logits { get; set; } Per-point class logits [N, numClasses].
TimeSpan InferenceTime { get; set; } Inference time.
int NumClasses { get; set; } Number of classes.
int NumPoints { get; set; } Number of points in the input cloud.