Chicken Disease Classifier

A CNN-based image classification system for identifying chicken diseases using deep learning and computer vision techniques. This project uses DVC (Data Version Control) for pipeline management and reproducibility.

Features

• Deep Learning: Uses a CNN (VGG16 based) for image classification.
• DVC Pipeline: Automated pipeline for data ingestion, model definition, training, and evaluation.
• Web Interface: User-friendly web app built with FastAPI and Bootstrap for easy interaction.
• Reproducibility: Global random seed (42) and GPU memory growth configuration for consistent results.
• Experiment Tracking: Tracks parameters (epochs, batch size, etc.) and metrics (accuracy, loss).
• AWS S3 Storage: Hybrid local/cloud artifact storage with automatic model uploads.
• CloudWatch Monitoring: Real-time training & prediction logging with custom metrics.

AWS Integration

This project supports AWS S3 for artifact storage and AWS CloudWatch for monitoring and observability.

Architecture

graph LR
    subgraph Local["🖥️ Local Environment"]
        TP["Training Pipeline"]
        FA["FastAPI App /predict"]
        CWC["CloudWatch Callback"]
    end

    subgraph AWS["☁️ AWS Cloud"]
        S3["S3 Bucket\n(Model, Data, Logs)"]
        CWL["CloudWatch Logs\n(Predictions, Errors)"]
        CWM["CloudWatch Metrics\n(Loss, Accuracy)"]
    end

    TP -->|upload| S3
    FA -->|logs| CWL
    CWC -->|metrics| CWM

    style AWS fill:#FF9900,color:#fff
    style Local fill:#232F3E,color:#fff

Loading

Setup

1. Copy .env.example to .env and fill in your AWS credentials:

   cp .env.example .env

2. Set STORAGE_MODE=s3 and ENABLE_CLOUDWATCH=true in .env.
3. Create your S3 bucket and CloudWatch log group in AWS Console.

CloudWatch Metrics Tracked

Metric	Description
TrainingLoss	Loss per epoch during training
TrainingAccuracy	Accuracy per epoch during training
PredictionCount	Count per prediction (with Class dimension)
PredictionLatency	Time taken per prediction (seconds)
DiseaseDetected	Binary flag for disease detection

CloudWatch Recordings

Training Logs in CloudWatch:

Prediction Logs in CloudWatch:

System Configuration

The pipeline has been tested on the following configuration:

• OS: macOS 26.1
• Model: MacBook Pro
• Chip: Apple M3 Pro
• Cores: 11 (5 performance and 6 efficiency)
• Memory: 18 GB

Installation

1. Clone the repository:

   git clone https://github.com/neehanthreddym/chicken_disease_clf.git
   cd chicken_disease_clf

2. Install dependencies:

   pip install -r requirements.txt

Usage (DVC Pipeline)

To run the entire machine learning pipeline (Data Ingestion -> Model Definition -> Training -> Evaluation):

dvc repro

This will check for changes in dependencies and only run the necessary stages.

Pipeline Stages

1. Data Ingestion (stage01_data_ingestion.py): Downloads and extracts the dataset.
2. Model Definition (stage02_model_definition.py): Prepares the VGG16 base model.
3. Model Training (stage03_training.py): Trains the model with augmented data.
4. Model Evaluation (stage04_evaluation.py): Evaluates the trained model and saves scores.

Web Application

The project includes a FastAPI-based web interface to easily classify images.

1. Start the application:

   python app.py

2. Open your browser and navigate to http://localhost:8000.

Reproducibility

• Random Seed: A global seed of 42 is set for Python, NumPy, and TensorFlow to ensure reproducible training runs.
• GPU Config: TensorFlow GPU memory growth is enabled to prevent allocation errors.

Development Workflow

Follow these steps when making changes to the project:

1. Update config.yaml: Modify system configuration settings (paths, URLs).
2. Update secrets.yaml (Optional): Add sensitive credentials like API keys.
3. Update params.yaml: Adjust parameters for model training/testing (Epochs, Batch Size, etc.).
4. Update the Entity: Modify data entities (dataclasses) in src/entity for accurate input/output.
5. Update the Configuration Manager: Adjust src/config/configuration.py to handle new configs.
6. Update the Components: Modify or add components in src/components.
7. Update the Pipeline: Update the processing steps in src/pipeline.
8. Update main.py: Modify the main script if necessary.
9. Update dvc.yaml: Update stage dependencies and outputs if the workflow changes.