PlantNet/plantnet-300k-v2-cnxt

Pl@ntNet-300K-v2 — DINOv3 ConvNeXt-Tiny

This is a 1,000-species plant classifier obtained by fine-tuning facebook/dinov3-convnext-tiny-pretrain-lvd1689m on Pl@ntNet-300K-v2.

Property	Value
Architecture	DINOv3 ConvNeXt-Tiny
Classes	1,000
Embedding dimension	768
Best validation accuracy	0.8248
Test accuracy	0.8216
Input size	224 × 224 RGB

Usage

The repository stores the fine-tuned encoder and classifier in model.pth. The included loader returns both image embeddings and classifier logits.

import sys

import pandas as pd
import torch
from huggingface_hub import snapshot_download
from PIL import Image

model_id = "PlantNet/plantnet-300k-v2-cnxt"
model_dir = snapshot_download(model_id)
sys.path.insert(0, model_dir)

from model import load_model
from transforms import ValTransform

device = "cuda" if torch.cuda.is_available() else "cpu"
model = load_model("cnxt", f"{model_dir}/model.pth", device=device)
transform = ValTransform()

image = transform(Image.open(f"{model_dir}/example.jpg").convert("RGB"))
with torch.inference_mode():
    embeddings, logits = model(image.unsqueeze(0).to(device))

species_id = logits.argmax(dim=1).item()
metadata = pd.read_csv(f"{model_dir}/species_metadata.csv")
species_name = metadata.loc[
    metadata["species_id"] == species_id, "full_species"
].item()
print(species_name, embeddings.shape)

The DINOv3 base model is license-gated. Accept its license and authenticate with Hugging Face before the first load.

Preprocessing

Images are resized/cropped to 224 × 224 and normalized with the Pl@ntNet statistics used during fine-tuning:

mean = (0.4425, 0.4695, 0.3266)
std  = (0.2353, 0.2219, 0.2325)

Citation

@inproceedings{plantnet-300k,
  author    = {Garcin, Camille and Joly, Alexis and Bonnet, Pierre and
               Lombardo, Jean-Christophe and Affouard, Antoine and
               Chouet, Mathias and Servajean, Maximilien and
               Lorieul, Titouan and Salmon, Joseph},
  booktitle = {NeurIPS Datasets and Benchmarks 2021},
  title     = {{Pl@ntNet-300K}: a plant image dataset with high label
               ambiguity and a long-tailed distribution},
  year      = {2021}
}

@misc{moummad2026,
  title         = {Energy-Efficient Plant Monitoring via Knowledge Distillation},
  author        = {Ilyass Moummad and Reda Bensaid and Kawtar Zaher and
                   Hervé Goëau and Jean-Christophe Lombardo and Joseph Salmon and
                   Pierre Bonnet and Alexis Joly},
  year          = {2026},
  eprint        = {2604.27178},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2604.27178}
}