GiladWeinberger/grape-leaf-5deg

Grape Leaf Downy Mildew Regressor (5deg variant)

Convolutional neural network that estimates the percentage of downy mildew on grape leaf images (0–100%).

Model details

Property	Value
Task	Image regression
Variant	5deg (5-degree rotation augmentation)
Input	RGB image, resized to 224×224
Output	Single scalar (affected area %)
Architecture	3-layer CNN + 2-layer MLP regressor

Preprocessing

transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
])

No ImageNet normalization is applied.

Usage (local PyTorch)

import torch
from PIL import Image
from torchvision import transforms

# Copy model.py from this repo or from the Django app ml/model.py
from model import GrapeLeafRegressor

model = GrapeLeafRegressor()
model.load_state_dict(torch.load("grape_leaf_model_5deg.pth", map_location="cpu"))
model.eval()

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
])

image = Image.open("leaf.jpg").convert("RGB")
tensor = transform(image).unsqueeze(0)

with torch.no_grad():
    prediction = model(tensor).item()

prediction = max(0, min(100, prediction))
print(f"Estimated downy mildew: {prediction:.1f}%")

Inference API (HF Space)

Deploy the companion Space from deploy/huggingface/space/ for a Gradio HTTP API used by the Vercel-hosted Django app.

Training

• Dataset: data/final_images_5deg (train / validation / test split)
• Loss: MSE, Optimizer: Adam (lr=0.001)
• Early stopping on validation loss (patience 5)
• Train locally or in Google Colab: python train_model.py --variant 5deg

Limitations

• Trained on augmented lab/field leaf images; performance may drop on very different lighting or species.
• Filename-based labels during training; not suitable for unlabeled arbitrary images without retraining.

Citation

Machine Learning Project — Shamir Institute (Dr. Lior Gur), with Daniel Kusai.