DINOv3Options

Options & Configuration · AiDotNet.VisionLanguage.Encoders

Configuration options for DINOv3, the next evolution of self-supervised vision encoders.

Inheritance: ModelOptions → VisionEncoderOptions → DINOv3Options

How It Works

DINOv3 (Meta, 2025) scales self-supervised ViT to 7B parameters trained on 1.7B images, outperforming SigLIP 2 on most vision benchmarks. It introduces improved training recipes with enhanced data augmentation and longer training schedules.

Constructors

Constructor Summary
DINOv3Options() Initializes a new instance with default values.
DINOv3Options(DINOv3Options 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.
bool UseSwiGLU { get; set; } Gets or sets whether to use SwiGLU activation in the FFN.
double IBOTMaskRatio { get; set; } Gets or sets the iBOT mask ratio.
int DINOHeadDim { get; set; } Gets or sets the self-supervised head dimension.
int NumRegisterTokens { get; set; } Gets or sets the number of register tokens.