DetectionBackboneOptions
Options & Configuration · AiDotNet.Models.Options
Options shared by the detection backbones in src/ComputerVision/Detection/Backbones —
the feature extractors (ResNet, EfficientNet, SwinTransformer, CSPDarknet) that a detector sits
on top of.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → DetectionBackboneOptions
Known derived types: CSPDarknetOptions, EfficientNetBackboneOptions, ResNetBackboneOptions, SwinTransformerOptions, VGG16BNBackboneOptions, VGG16BackboneOptions, YoloBackboneOptions
For Beginners
A backbone is the part of a vision model that turns an image into a stack of increasingly abstract feature maps. Detectors and segmenters bolt their own heads onto one of these rather than starting from pixels themselves, which is why the same four backbones appear across many different models.
How It Works
Only InChannels is declared here: it is the one knob all four share and all four
called by that name. Each backbone’s Variant names its own enum
(ResNetVariant, SwinVariant, …) so there is no common type to declare, and it
stays on the leaf — the same reasoning that kept ModelSize off
SegmentationModelOptions.
These classes are named …BackboneOptions rather than ResNetOptions /
EfficientNetOptions because those names are already taken by the options for the
standalone ResNetNetwork and EfficientNetNetwork classifiers, which are different
types with their own configuration.
Properties
| Property | Summary |
|---|---|
int InChannels { get; set; } |
Gets or sets the number of channels in the input image. |