Plant Classification Model Web App

This project is a Flask-based web application that uses a PyTorch model to classify plant images into one of ten predefined categories. It employs a pre-trained Vision Transformer (ViT) model fine-tuned for plant identification.

---

Screenshots

• Main Interface:

---

Features

• Image Classification: Classifies uploaded plant images into one of the following categories:
  • Aloe Vera
  • Areca Palm
  • Boston Fern
  • Chinese Evergreen
  • Dracaena
  • Money Tree
  • Peace Lily
  • Rubber Plant
  • Snake Plant
  • ZZ Plant
• Confidence Score: Calculates the confidence score for the predicted class.
• Interactive Interface: Simple web-based UI for uploading images and viewing predictions.

---

Requirements

Software

• Python 3.8 or above
• Flask
• PyTorch
• Transformers
• Torchvision
• PIL (Pillow)

Install Dependencies

Install the required libraries using:

pip install flask torch torchvision transformers pillow

---

Model

The model uses a fine-tuned Vision Transformer (ViT) from the Hugging Face Transformers library:

• Pretrained Model: umutbozdag/plant-identity
• Framework: PyTorch
• Input Size: Resized to 224x224 pixels
• Normalization: Standard ImageNet normalization.

---

Usage

1. Clone the repository:

   git clone https://github.com/your-username/plant-classification.git
   cd plant-classification

2. Place the model weights (model.pth) in the root directory.

3. Start the Flask application:

   python app.py

4. Open your browser and navigate to:

   http://127.0.0.1:5000/
   

5. Upload a plant image to classify it.

---

File Structure

• app.py: Main Flask application for handling requests and predictions.
• templates/index.html: Front-end page for uploading images and displaying results.
• model.pth: Saved PyTorch model weights (place in the root directory).
• static/: Folder for any static assets (e.g., CSS, images).

---

Example Workflow

1. User uploads a plant image via the web interface.
2. The image is processed using PyTorch transforms:
   • Resized to 224x224 pixels.
   • Normalized with ImageNet mean and standard deviation.
3. The pre-trained ViT model predicts the plant category.
4. The predicted class is displayed on the web page.

---

Notes

• Ensure the model.pth file matches the model architecture used in app.py.
• Use a CUDA-compatible GPU for faster inference, or the application will default to CPU.
• The class names are customizable if the model is retrained for additional plants.