TigranBoyakhchyan/plant-disease-classifier

Plant Disease Classifier

Species-invariant plant disease classification. Given an image of a plant leaf, predicts the disease regardless of the host plant species.

Model

• Backbone: ConvNeXt-Tiny (27.85M parameters)
• Input: 224x224 RGB images
• Output: 39 disease classes
• Training: Fine-tuned from ImageNet pretraining with aggressive augmentation (RandAugment + random erasing)

Results

Metric	Value
Standard validation accuracy	0.628
Holdout (leave-one-species-out) validation accuracy	0.254

The holdout split holds out specific (species, disease) pairs from training, so the model is evaluated on unseen species-disease combinations. This is a stronger test of whether the model learned disease features vs. species features.

Files

• final_model.pth — PyTorch checkpoint (model weights + metadata)
• labels.json — class index mapping and metadata

Usage

Install dependencies and load:

import torch, timm
from torchvision import transforms
from PIL import Image
from huggingface_hub import hf_hub_download

path = hf_hub_download(repo_id="TigranBoyakhchyan/plant-disease-classifier", filename="final_model.pth")
ckpt = torch.load(path, map_location="cpu", weights_only=False)

model = timm.create_model(ckpt["backbone"], pretrained=False, num_classes=ckpt["num_classes"])
model.load_state_dict(ckpt["model_state_dict"])
model.eval()

idx_to_disease = {v: k for k, v in ckpt["disease_to_idx"].items()}

tfm = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(ckpt["img_size"]),
    transforms.ToTensor(),
    transforms.Normalize(ckpt["mean"], ckpt["std"]),
])

img = Image.open("leaf.jpg").convert("RGB")
x = tfm(img).unsqueeze(0)
with torch.no_grad():
    logits = model(x)
pred = idx_to_disease[logits.argmax(1).item()]
print(pred)