README.md

metadata

license: mit
title: Plant Disease Detection App
sdk: gradio
emoji: 🚀
colorFrom: red
colorTo: blue

5CCSAGAP: AI/Robotics Group Project - Plant Disease Detection

This repository contains the source code for the "Plant Disease Detector" application, developed for the 5CCSAGAP module.

1. Project Overview

The goal of this project is to build an AI system capable of detecting plant diseases from leaf images. The application is built with a Python backend using PyTorch and is deployed with a Gradio user interface. The system supports two models: a custom-built CNN and a high-performance model based on ResNet18 transfer learning.

Team 4 Members:

• Oguzhan Cagirir
• Janit Bhardwaj
• Hristina Georgieva
• Hissan Omar
• Kasim Morsel
• Mark Soltyk

2. Running the Application

This guide is for running the Gradio application locally to demonstrate its functionality.

Prerequisites

• Python 3.10+
• An active internet connection (for downloading model weights on first run).
• Git LFS is not required as large files are hosted externally.

Setup Instructions

1. Clone the repository:

   git clone https://github.kcl.ac.uk/k23136072/Small-group-project.git
   cd Small-group-project
   

2. Create and activate a Python virtual environment:

   On Windows:

   python -m venv .venv
   .venv\Scripts\activate
   

   On Linux/macOS:

   python3 -m venv .venv
   source .venv/bin/activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

Launching the Gradio UI

Once the setup is complete, launch the application:

python app.py

The application will start and provide a local URL (e.g., http://127.0.0.1:7860). Open this URL in your web browser.

Note on First Run: The first time you make a prediction with a specific model (e.g., "ResNet18"), the application will automatically download the required model weights (~45MB) from Hugging Face Hub. This may take a moment. Subsequent runs will use the cached local copy.

3. Repository Structure

• app.py: The main Gradio application entry point.
• src/: Contains all core source code.
  • src/inference.py: The core API for running model predictions. Handles model downloading and preprocessing.
  • src/models/: Contains the PyTorch model definitions (cnn_model.py, resnet18_finetune.py).
  • src/DataLoader/: Contains the data loading and processing logic used for training.
  • src/train/: Contains the training (train.py) and evaluation (evaluate.py) scripts.
  • src/utils/: Configuration and helper utilities.
• configs/: Contains the YAML configuration file for training.
• requirements.txt: Project dependencies.
• process_dataset.py: Utility script to download and prepare the dataset from Hugging Face.
• README.md: This file.

4. Model Information

The application supports two models, selectable via the UI dropdown:

1. ResNet18 (Default): A high-performance model using transfer learning from a pretrained ResNet18. Achieves ~96% accuracy on the test set.
2. CNN: A custom-built baseline CNN, trained from scratch. Achieves ~78% accuracy on the test set.

Model weights are hosted on Hugging Face Hub at MZaik/Plant_Disease_Detection and are downloaded automatically by the application.

5. Training and Evaluation (Advanced)

To reproduce the training or evaluation runs, the repository is integrated with ClearML for experiment tracking.

Training:

1. Configure configs/train_cnn.yaml (set model_type to resnet18 or cnn).
2. Run python -m src.train.train.
   • This will automatically trigger data processing if the dataset is missing.
   • It will execute on the configured ClearML server if credentials are present, or locally if not.

Evaluation:

1. Obtain a Model Task ID from a successful training run on ClearML.
2. Run python -m src.evaluate --model_type resnet18 --task_id [TASK_ID].
   • This generates the confusion matrix and classification report artifacts.

---