🌿 Plant Disease Detector

Upload a clear photo of a single leaf. The model recognises 38 plant/disease classes. Everything runs in your browser; your photo is not uploaded.

Click to choose a photo, or drop one here

Loading model (about 17 MB)...

Result

Choose a photo or click an example below.

Model: EfficientNetB0 transfer learning (ImageNet weights, two-phase fine-tuning), 38 plant/disease classes.
Accuracy: 99.08% on the held-out test split of the dataset. That split comes from the same source as the training data, so it is not a measure of real-world accuracy.
Limitations: trained on PlantVillage lab-style images (single leaf, plain background), so field photos may be less accurate. In our own check, accuracy dropped on blurry photos.
This page: the same trained Keras model converted to ONNX and run in your browser with onnxruntime-web. The Python/Gradio version of this app runs locally. The model was trained in a Google Colab notebook and its experiments and model registry are tracked with MLflow.
Credits: Utsav Singh and Mansi Pathak.