BiFPN<T>
Models & Types · AiDotNet.ComputerVision.Detection.Necks
Bidirectional Feature Pyramid Network (BiFPN) with weighted feature fusion.
Inheritance: ModelBase<T, Tensor<T>, Tensor<T>> → NeckBase<T> → BiFPN<T>
For Beginners
BiFPN improves upon FPN and PANet by using learnable weights for feature fusion. Instead of simply adding features, it learns how much each input feature should contribute to the fused output.
How It Works
Key features:
- Bidirectional (top-down + bottom-up) feature flow
- Learnable weights for weighted feature fusion
- Fast normalized fusion with softmax
- Used in EfficientDet for state-of-the-art detection
Reference: Tan et al., “EfficientDet: Scalable and Efficient Object Detection”, CVPR 2020
Constructors
| Constructor | Summary |
|---|---|
BiFPN(NeckConfig config) |
Creates BiFPN from a configuration object. |
BiFPN(int[] inputChannels, int outputChannels, int numRepeats) |
Creates a new Bidirectional Feature Pyramid Network. |
Properties
| Property | Summary |
|---|---|
int NumLevels { get; } |
|
int OutputChannels { get; } |
|
string Name { get; } |
Methods
| Method | Summary |
|---|---|
List<Tensor<T>> Forward(List<Tensor<T>> features) |
|
long GetParameterCount() |
|
void ReadParameters(BinaryReader reader) |
|
void WriteParameters(BinaryWriter writer) |