Skip to content

README.md

Latest commit

 

History

History
77 lines (51 loc) · 2.22 KB

README.md

File metadata and controls

77 lines (51 loc) · 2.22 KB

PlantDoc model reproduction

Classification setup from PlantDoc: A Dataset for Visual Plant Disease Detection (CoDS-COMAD 2020).
Dataset: pratikkayal/PlantDoc-Dataset.

1. Dataset

Clone the dataset so that train/ and test/ live under data/PlantDoc-Dataset/:

mkdir -p data && cd data && git clone https://github.com/pratikkayal/PlantDoc-Dataset.git
  • data/PlantDoc-Dataset/train/ — one subfolder per class
  • data/PlantDoc-Dataset/test/ — same class subfolders

2. Install dependencies

pip install -r requirements.txt

3. Train

python -m plantdoc.train --model VGG16 --epochs 50

Other models: InceptionV3, InceptionResNetV2:

python -m plantdoc.train --model InceptionResNetV2 --epochs 50

Checkpoints are saved under outputs/ (e.g. outputs/vgg16_plantdoc.keras).

Train in the cloud

Upload notebooks/plantdoc_train_colab.ipynb to Google Colab, set runtime to T4 GPU, and run all cells. Download the .keras file and the .weights.h5 file from the notebook.

Using the trained model (inference)

Put vgg16_plantdoc.weights.h5 in the project root. Optionally add class_names.json for readable labels.

python -m plantdoc.predict vgg16_plantdoc.weights.h5 path/to/leaf.jpg --classes class_names.json

Example output:

Prediction: Tomato leaf bacterial spot
Confidence: 87.32%

FastAPI backend

Put vgg16_plantdoc.weights.h5 and optionally class_names.json in the project root.

pip install -r requirements.txt
uvicorn main:app --reload
  • POST /predict — upload a leaf image (form field file). Returns label, confidence, all_classes.
  • GET /health — check that the model is loaded.

Example:

curl -X POST http://127.0.0.1:8000/predict -F "file=@data/PlantDoc-Dataset/test/Apple leaf/4120978-single-green-leaf-of-apple-tree.jpg"

Docs: http://127.0.0.1:8000/docs

Reference