Module dinov2reg4 

Implementation of the DINOv2 revision (4 regularization)

The DINOv2-reg4 model is a variant of DINOv2 that adds 4 regularization tokens to the original architecture. This implementation is specifically trained for plant species classification on the PlantCLEF2024 dataset with 7,806 classes.

• Paper. DINOv2: Learning Robust Visual Features without Supervision
• GH Repo

Example

# Download classes names and a plant picture to identify
# see candle/examples/dinov2reg4 for full code.

# Perform inference
cargo run \
  --example dinov2reg4 \
  --release -- \
  --image <orchid-file>

> Orchis simia Lam.       : 45.55%
> Orchis × bergonii Nanteuil: 9.80%
> Orchis italica Poir.    : 9.66%
> Orchis × angusticruris Franch.: 2.76%
> Orchis × bivonae Tod.   : 2.54%

Structs

DinoVisionTransformer

Functions

vit_base

vit_small