PlantNet-300K MobileNetV3-Small (v2 Improved)

Fine-tuned plant species classifier trained on 1,081 plant species from the PlantNet-300K dataset.

Model Details

Architecture: MobileNetV3-Small
Parameters: ~2.5M (10 MB ONNX fp32)
Input: 224×224 RGB images
Training Dataset: PlantNet-300K (306,406 training images, 1,081 species)
Version: v2 (improved hyperparameters)

Performance

v2 (This Model) — Improved

Metric	Train	Val	Test
Top-1 Accuracy	74.20%	75.56%	75.45% ⭐
Top-5 Accuracy	93.17%	—	93.81%
Loss	2.0298	—	1.9623

v1 (Previous)

Metric	Test
Top-1 Accuracy	73.89%
Top-5 Accuracy	91.86%

Improvement: +1.56 percentage points top-1 accuracy 🚀

Training Details

v2 Configuration (improved):

Optimizer: SGD with momentum 0.9, Nesterov=True
LR Schedule: Cosine annealing from 0.01 → 1e-5
Augmentation: TrivialAugment + RandomErasing
Class Balancing: WeightedRandomSampler
Label Smoothing: 0.1
Batch Size: 256
Total Epochs: 60 (Phase 1: 5 frozen, Phase 2: 55 full)

Key Improvements:

Better LR schedule (cosine vs fixed)
Stronger augmentation strategy
Class-weighted sampling for imbalanced data
Label smoothing for regularization

Usage

from PIL import Image
import torch
import torchvision.models as models
import torchvision.transforms as transforms

# Load model
model = models.mobilenet_v3_small(weights=None, num_classes=1081)
model.load_state_dict(torch.hub.load_state_dict_from_url(
    'https://huggingface.co/cpoisson/plantnet300k-mobilenetv3-small/resolve/main/mobilenetv3_small_v2.pth'
))
model.eval()

# Prepare image
transform = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]
    )
])

image = transform(Image.open('plant.jpg')).unsqueeze(0)

# Predict
with torch.no_grad():
    logits = model(image)
    top_k = torch.topk(logits, 5)
    probs = torch.softmax(logits, dim=1)
    
print(f"Top-1: {probs.max().item():.2%}")

Files

mobilenetv3_small_v2.pth — v2 PyTorch weights (best checkpoint from epoch 59)
mobilenetv3_small_v2.onnx — v2 ONNX fp32 export (10 MB, browser-compatible)

Dataset

PlantNet-300K: 306,406 high-quality plant images across 1,081 species, collected via the PlantNet mobile app.

Train: 245,402 images
Val: 29,892 images
Test: 31,112 images

Download Dataset

Related Resources

Gradio Demo: cpoisson/plantnet300k
Offline Browser Demo: plantnet300k-offline
v1 Model: plantnet300k-mobilenetv3-small
ResNet18 v1: plantnet300k-resnet18

Citation

If you use this model, please cite:

@article{plantnet2017,
  title={PlantNet: A Large-Scale Continuous Ecosystem for Plant Image Classification},
  author={Cole et al.},
  year={2017}
}

License

OpenRAIL — Free for research and commercial use.

Downloads last month: -; Downloads are not tracked for this model. How to track

cpoisson
/

plantnet300k-mobilenetv3-small