PointConvolutionLayer<T>
Layers ยท AiDotNet.PointCloud.Layers
Implements a convolution layer specifically designed for point cloud data.
Inheritance: LayerBase<T> โ PointConvolutionLayer<T>
Implements: IShapeContract
How It Works
For Beginners: Unlike regular convolutions for images, point cloud convolutions work on unordered 3D points.
Key differences from image convolutions:
- Images have regular grid structure (pixels in rows/columns)
- Point clouds are unordered sets of 3D coordinates
- Must be invariant to point order (permutation invariant)
- Must handle varying number of points
This layer applies a shared per-point linear map (a 1x1 convolution over points): output[p] = activation(W^T x[p] + b), learned weights that work regardless of point order.
Applications:
- Feature extraction from local 3D geometry
- Learning shape patterns in point clouds
- Building blocks for PointNet / DGCNN-style architectures
Constructors
| Constructor | Summary |
|---|---|
PointConvolutionLayer(int inputChannels, int outputChannels, IActivationFunction<T> activation) |
Initializes a new instance of the PointConvolutionLayer class. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<string, string> GetParameterRoles() |
Returns parameter roles for per-role optimizer configuration (e.g., weight decay exemption for biases). |
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
|
IReadOnlyList<Tensor<T>> GetTrainableParameters() |
Returns all trainable parameter tensors marked with [TrainableParameter]. |
void ClearGradients() |
|
void ResetState() |
|
void SetTrainableParameters(IReadOnlyList<Tensor<T>> parameters) |
Replaces trainable parameter tensors (e.g., with ParameterBuffer views). |
void ZeroGrad() |
Clears all gradient fields discovered by convention ({paramName}Gradient). |