EdManZoeTech/plant-disease-detection

Plant Disease Detection Dataset (23-Class Leaf Images)

A curated dataset of labeled plant-leaf images designed to support plant disease classification. This dataset is suitable for training image classification models, fine-tuning vision foundation models, and building agricultural decision-support and education tools.

---

🌱 Dataset Overview

• Total Images: 35,725 (256×256 JPEG crops)
• Classes: 23 disease/healthy categories across 5 plant species
• Species: Apple, Corn (maize), Pepper bell, Potato, Tomato
• Healthy / Diseased: 7,088 / 28,637
• Total Size: ~556 MB
• Labels: English, stored in metadata.csv (23-column schema)

---

📦 Dataset Summary

This dataset contains labeled leaf images paired with rich metadata, making it ideal for:

• 🌿 Plant disease classification model training
• 🔬 Transfer learning with vision models (CNNs, ViTs)
• 🚜 Agricultural decision-support and crop-health applications
• 🧠 Few-shot / fine-tuning research on imbalanced biological data
• 🎓 Education in plant pathology and applied machine learning

---

✂️ Splits

Split	Images	Size (bytes)
train	28,579	444,918,266
validation	3,573	55,746,956
test	3,573	55,806,484
total	35,725	556,471,706

---

🏷️ Classes (23)

Apple___Apple_scab, Apple___Black_rot, Apple___Cedar_apple_rust, Apple___healthy, Corn_(maize)___Cercospora_leaf_spot Gray_leaf_spot, Corn_(maize)___Common_rust_, Corn_(maize)___Northern_Leaf_Blight, Corn_(maize)___healthy, Pepper__bell___Bacterial_spot, Pepper__bell___healthy, Potato___Early_blight, Potato___Late_blight, Potato___healthy, Tomato_Bacterial_spot, Tomato_Early_blight, Tomato_Late_blight, Tomato_Leaf_Mold, Tomato_Septoria_leaf_spot, Tomato_Spider_mites_Two_spotted_spider_mite, Tomato__Target_Spot, Tomato__Tomato_YellowLeaf__Curl_Virus, Tomato__Tomato_mosaic_virus, Tomato_healthy

Class folder names are intentional and must not be normalized — they are the shared contract between the dataset, the Leaf Ledger client, and the server.

---

🖼️ Image Access

• metadata.csv contains one row per image (23-column schema); the image_path column holds the direct R2 URL of each image.
• Local copies live under dataset/<class folder>/<file>.jpg.

---

📜 License

This dataset is distributed under the MIT License. Additional licensing options: Apache 2.0 and Creative Commons Attribution.

Please ensure compliance with relevant terms when using or modifying the dataset.

---

📖 Citation

If you use this dataset in your research or projects, please cite:

@dataset{edman_plant_disease_detection_2026,
  title     = {Plant Disease Detection Dataset (23-Class Leaf Images)},
  author    = {EdManZoeTech},
  year      = {2026},
  url       = {https://huggingface.co/datasets/EdManZoeTech/plant-disease-detection}
}

---

🤝 Contributing

Contributions are welcome! Here's how to help:

1. Fork this repository
2. Create a feature branch
3. Submit a pull request with a clear explanation

New field-collected images enter through the Leaf Ledger client and are validated against the shared 23-class taxonomy.

---

📬 Contact

For questions, feedback, or collaboration inquiries, please reach out via the Hugging Face hub or contact EdManZoeTech directly.

---

🙏 Acknowledgments

• The original PlantVillage dataset and its authors (research basis of the labeled leaf images)
• The Leaf Ledger field contributors
• The plant-pathology open-data community
• Hugging Face for the dataset hosting infrastructure

---

🔔 This dataset is part of ongoing efforts to make plant-disease detection accessible through open data and open-source tooling.

---