MaxPoolingLayer<T>
Layers · AiDotNet.PointCloud.Layers
Implements global max pooling for point clouds to extract global features.
Inheritance: LayerBase<T> → MaxPoolingLayer<T>
Implements: IShapeContract
How It Works
For Beginners: Max pooling takes the maximum value across all points for each feature channel.
How it works:
- Input: N points, each with C features [N, C]
- Operation: For each feature channel, find the maximum value across all N points
- Output: A single vector of C features [1, C]
Why it’s useful:
- Creates a global representation of the entire point cloud
- Achieves permutation invariance (order of points doesn’t matter)
- Reduces dimensionality from many points to one feature vector
Example:
- Input: 1024 points with 64 features each = [1024, 64]
- Max pooling across points
- Output: 1 global feature vector with 64 features = [1, 64]
This is a key component in PointNet for making the network invariant to point order.
Constructors
| Constructor | Summary |
|---|---|
MaxPoolingLayer(int numFeatures) |
Initializes a new instance of the MaxPoolingLayer class. |
Properties
| Property | Summary |
|---|---|
bool SupportsTraining { get; } |
Methods
| Method | Summary |
|---|---|
IReadOnlyList<OutputAxisContract> OutputAxesFor(int inputRank) |
|
void ClearGradients() |
|
void ResetState() |