DINOOptions<T>
Options & Configuration · AiDotNet.Models.Options
Hyperparameters of DINO (Zhang et al., “DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection”, ICLR 2023). The defaults are the paper’s DINO-4scale configuration.
Inheritance: ModelOptions → ObjectDetectionOptions<T> → DINOOptions<T>
For Beginners
The defaults reproduce the published model; change them only to experiment.
How It Works
Size selects the paper’s backbone: Nano, Small and Medium use
ResNet-50 at four scales, Large uses Swin-L at four scales, and XLarge uses Swin-L at five scales. The
transformer is the same for every size, as in the paper.
Constructors
| Constructor | Summary |
|---|---|
DINOOptions() |
Creates the options with the paper’s DINO-4scale values. |
Properties
| Property | Summary |
|---|---|
double BoxNoiseScale { get; set; } |
Box noise scale (lambda). |
double LabelNoiseRatio { get; set; } |
Label noise ratio. |
double PositionalTemperature { get; set; } |
Temperature of the 2-D sine positional encoding. |
int DenoisingQueries { get; set; } |
Contrastive-denoising query budget (dn_number). |
int FeedForwardDimension { get; set; } |
Feed-forward width in the encoder and decoder layers. |
int HiddenDimension { get; set; } |
Transformer width (d_model). |
int NumDecoderLayers { get; set; } |
Decoder layers. |
int NumEncoderLayers { get; set; } |
Deformable encoder layers. |
int NumHeads { get; set; } |
Attention heads in every attention block. |
int NumQueries { get; set; } |
Object queries selected from the encoder (mixed query selection). |
int NumSamplingPoints { get; set; } |
Sampling points per head and level in deformable attention. |