ResNet<T>
Models & Types · AiDotNet.ComputerVision.Detection.Backbones
ResNet backbone network for feature extraction.
Inheritance: NeuralNetworkBase<T> → ResNet<T>
Implements: IDetectionBackbone<T>
For Beginners
ResNet (Residual Network) is a foundational architecture that introduced skip connections to enable training of very deep networks. It’s widely used as a backbone for detection models like Faster R-CNN.
How It Works
Key features:
- Residual blocks with skip connections prevent gradient vanishing
- Multiple variants: ResNet-18, 34, 50, 101, 152
- Bottleneck blocks (3 convolutions) for deeper networks
Reference: He et al., “Deep Residual Learning for Image Recognition”, CVPR 2016
Constructors
| Constructor | Summary |
|---|---|
ResNet(ResNetBackboneOptions options) |
Creates a new ResNet backbone: the classifier’s trunk (torchvision’s ResNet without its pooling and fully connected layer), producing the C2–C5 feature pyramid. |
Properties
| Property | Summary |
|---|---|
IReadOnlyList<int> OutputChannels { get; } |
|
IReadOnlyList<int> Strides { get; } |
|
bool IsFrozen { get; set; } |
|
string Name { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T> input) |
Returns the stem output and each residual stage’s output (the C2..C5 pyramid), keyed by name. |
IFullModel<T, Tensor<T>, Tensor<T>> WithParameters(Vector<T> parameters) |
|
IReadOnlyList<Tensor<T>> GetFeatureMaps(Tensor<T> input) |
|
List<Tensor<T>> ExtractFeatures(Tensor<T> input) |
|
ModelMetadata<T> GetModelMetadata() |
|
ValueTuple<int, int> GetExpectedInputSize() |
|
void Freeze() |
|
void ReadParameters(BinaryReader reader) |
|
void Unfreeze() |
|
void WriteParameters(BinaryWriter writer) |