French trees — Pl@ntNet ViT-B/14 fine-tune (accurate)
Pl@ntNet's PlantCLEF 2024 model (DINOv2 ViT-B/14, 7,806 species of south-western Europe) fine-tuned on 71 common trees of France plus an other plant class. Best single-model accuracy in our tests; use it for photos, and the small sibling model for live video.
Sibling model: Maximaxou2/french-trees-vit-small — a 4× smaller model distilled from this one, for live/mobile use.
Usage
import json, timm, torch
from huggingface_hub import hf_hub_download
from PIL import Image
model = timm.create_model("hf-hub:Maximaxou2/french-trees-plantnet-ft", pretrained=True).eval()
transform = timm.data.create_transform(**timm.data.resolve_data_config({}, model=model))
labels = json.load(open(hf_hub_download("Maximaxou2/french-trees-plantnet-ft", "labels.json"), encoding="utf8"))
temperature = json.load(open(hf_hub_download("Maximaxou2/french-trees-plantnet-ft", "config.json")))["temperature"]
x = transform(Image.open("photo.jpg").convert("RGB")).unsqueeze(0)
with torch.no_grad():
probs = (model(x) / temperature).softmax(-1)[0]
for p, i in zip(*probs.topk(3)):
print(f"{p:.1%} {labels[i]['latin']} ({labels[i]['fr']})")
Divide logits by temperature (fitted on the validation split) for calibrated probabilities.
Class 71 (other) means "probably not one of the 71 trees".
ONNX: model.onnx — fp16 weights, fp32 input image [1, 3, 448, 448]
(RGB, center square crop, ImageNet mean/std), output logits [1, 72]. Runs with ONNX Runtime, incl. Android
(CPU/XNNPACK or Qualcomm NPU via QNN).
Results
Test set: 7,090 photos of the 71 trees (100 iNaturalist observations per species, 90 for Thuja) observed in 2025–2026 by people absent from the training data, so neither the base model nor this one saw them; plus 1,000 photos of other plants, half of them close relatives of the 71 (other oaks, maples, pines, Tilia × europaea, oriental plane…).
| Model | Top-1 (among the 71) | Open-set accuracy on trees | Other plants rejected |
|---|---|---|---|
| Pl@ntNet PlantCLEF 2024, zero-shot (original) | 89.8 | 69.0 | 90.4 |
| This model | 95.97 | 94.89 | 88.5 |
- Top-3 among the 71: 98.67 %. Calibration error (ECE): 0.54 %.
- When the calibrated confidence is ≥ 90 % (92.3 % of tree photos), it is right 98.66 % of the time.
- By photo type (auto-tagged, approximate): flower 98.2 · leaf 97.9 · fruit 97.0 · twig or bud 96.6 · needles or cones 96.1 · whole tree 89.2 · bark 88.9.
- Most frequent confusions: Tilia cordata → Tilia platyphyllos (16), Tilia platyphyllos → Tilia cordata (14), Quercus robur → Quercus petraea (9), Pinus pinea → Pinus pinaster (8), Pinus sylvestris → Pinus nigra (7), Pinus nigra → Pinus pinaster (6). These pairs are hard from a single photo even for botanists, and iNaturalist labels for them are themselves less reliable.
Size and speed: 86 M parameters, 448 px input; ≈ 600 img/s on an RTX 5090 (fp16, batch 64).
Training data
iNaturalist research-grade observations with CC-licensed photos (iNaturalist open-data bucket), France first, the rest of Europe for the less-observed species: up to 700 observations (≤ 2 photos each) per species, plus ~37k photos of ~5,000 other plant species for the other class. Observer-disjoint train / validation / test splits. Licenses of training photos: 81 % CC BY-NC, 12 % CC BY, 4 % CC0, the rest other CC variants.
Full fine-tune at 448 px, 8 epochs (AdamW, layer-wise LR decay 0.75, label smoothing 0.1, class-balanced loss), then 2 more epochs after adding 32k extra other plant photos. The new head was initialised from Pl@ntNet's own classifier rows for these species.
Limitations
- Only 71 species. Anything else should come out as
other, but ~11 % of lookalike plants are still named as one of the 71 — treat low-confidence answers as "check it". - Bark-only and whole-tree photos are clearly harder than leaves, flowers or fruits.
- Platanus × hispanica (London plane) is not in the base Pl@ntNet model; it was learned only during fine-tuning.
- Built for learning and curiosity, not for decisions that matter (e.g. edibility or toxicity).
License and credits
Weights: CC BY-NC 4.0 — most training photos are CC BY-NC. Base models: Pl@ntNet PlantCLEF 2024 ViT-B/14 (Apache-2.0, trained on Pl@ntNet data), DINOv2 (Meta). Photos: the iNaturalist community. Thank you to every observer.
Species
| Latin name | French name | Test accuracy (%) |
|---|---|---|
| Quercus robur | Chêne pédonculé | 83.0 |
| Quercus petraea | Chêne sessile | 93.0 |
| Quercus pubescens | Chêne pubescent | 97.0 |
| Quercus ilex | Chêne vert | 97.0 |
| Quercus suber | Chêne-liège | 98.0 |
| Quercus rubra | Chêne rouge d'Amérique | 99.0 |
| Quercus cerris | Chêne chevelu | 92.0 |
| Fagus sylvatica | Hêtre commun | 96.0 |
| Carpinus betulus | Charme commun | 91.0 |
| Castanea sativa | Châtaignier | 97.0 |
| Fraxinus excelsior | Frêne élevé | 96.0 |
| Fraxinus ornus | Frêne à fleurs | 96.0 |
| Acer pseudoplatanus | Érable sycomore | 97.0 |
| Acer platanoides | Érable plane | 97.0 |
| Acer campestre | Érable champêtre | 97.0 |
| Acer monspessulanum | Érable de Montpellier | 97.0 |
| Tilia cordata | Tilleul à petites feuilles | 79.0 |
| Tilia platyphyllos | Tilleul à grandes feuilles | 85.0 |
| Betula pendula | Bouleau verruqueux | 94.0 |
| Alnus glutinosa | Aulne glutineux | 99.0 |
| Corylus avellana | Noisetier | 100.0 |
| Juglans regia | Noyer commun | 96.0 |
| Prunus avium | Merisier | 97.0 |
| Prunus padus | Cerisier à grappes | 98.0 |
| Prunus mahaleb | Cerisier de Sainte-Lucie | 99.0 |
| Malus sylvestris | Pommier sauvage | 99.0 |
| Pyrus communis | Poirier commun | 97.0 |
| Sorbus aucuparia | Sorbier des oiseleurs | 98.0 |
| Sorbus domestica | Cormier | 94.0 |
| Sorbus torminalis | Alisier torminal | 100.0 |
| Sorbus aria | Alisier blanc | 99.0 |
| Crataegus monogyna | Aubépine monogyne | 98.0 |
| Robinia pseudoacacia | Robinier faux-acacia | 99.0 |
| Gleditsia triacanthos | Févier d'Amérique | 98.0 |
| Platanus × hispanica | Platane commun | 99.0 |
| Aesculus hippocastanum | Marronnier d'Inde | 100.0 |
| Ailanthus altissima | Ailante glanduleux | 98.0 |
| Celtis australis | Micocoulier de Provence | 99.0 |
| Populus nigra | Peuplier noir | 99.0 |
| Populus tremula | Peuplier tremble | 97.0 |
| Populus alba | Peuplier blanc | 98.0 |
| Salix alba | Saule blanc | 100.0 |
| Salix caprea | Saule marsault | 99.0 |
| Ulmus minor | Orme champêtre | 94.0 |
| Ulmus glabra | Orme de montagne | 96.0 |
| Ilex aquifolium | Houx | 99.0 |
| Buxus sempervirens | Buis | 100.0 |
| Cornus mas | Cornouiller mâle | 98.0 |
| Euonymus europaeus | Fusain d'Europe | 99.0 |
| Sambucus nigra | Sureau noir | 100.0 |
| Olea europaea | Olivier | 99.0 |
| Arbutus unedo | Arbousier | 98.0 |
| Laurus nobilis | Laurier-sauce | 98.0 |
| Prunus laurocerasus | Laurier-cerise | 96.0 |
| Ostrya carpinifolia | Charme-houblon | 99.0 |
| Pinus sylvestris | Pin sylvestre | 89.0 |
| Pinus pinaster | Pin maritime | 91.0 |
| Pinus nigra | Pin noir | 86.0 |
| Pinus halepensis | Pin d'Alep | 97.0 |
| Pinus pinea | Pin parasol | 84.0 |
| Pinus cembra | Pin cembro | 88.0 |
| Pinus uncinata | Pin à crochets | 94.0 |
| Abies alba | Sapin pectiné | 97.0 |
| Picea abies | Épicéa commun | 94.0 |
| Pseudotsuga menziesii | Douglas | 96.0 |
| Larix decidua | Mélèze d'Europe | 96.0 |
| Cedrus atlantica | Cèdre de l'Atlas | 97.0 |
| Taxus baccata | If commun | 94.0 |
| Juniperus communis | Genévrier commun | 99.0 |
| Cupressus sempervirens | Cyprès de Provence | 99.0 |
| Thuja occidentalis | Thuya du Canada | 95.6 |
other |
Autre plante | — |
- Downloads last month
- 13