DINOv2Options

Options & Configuration · AiDotNet.VisionLanguage.Encoders

Configuration options for the DINOv2 self-supervised vision encoder.

Inheritance: ModelOptions → VisionEncoderOptions → DINOv2Options

How It Works

DINOv2 (Oquab et al., 2024) trains ViT with self-supervised objectives (iBOT + DINO head) on 142M curated images (LVD-142M). It produces universal visual features without labels, achieving linear-probe results competitive with fine-tuned CLIP on many benchmarks.

Constructors

Constructor Summary
DINOv2Options() Initializes a new instance with default values.
DINOv2Options(DINOv2Options other) Initializes a new instance by copying from another instance.

Properties

Property Summary
bool UseRegisterTokens { get; set; } Gets or sets whether to use register tokens (DINOv2 with registers variant).
double IBOTMaskRatio { get; set; } Gets or sets the iBOT mask ratio for masked image modeling.
int DINOHeadDim { get; set; } Gets or sets the self-supervised head dimension for DINO loss.
int NumRegisterTokens { get; set; } Gets or sets the number of register tokens appended to the patch sequence.