Human Pose Classification - Deep Learning Project

A deep learning project that classifies human poses into three categories: Standing, Sitting, and Lying using transfer learning with MobileNetV2.

Project Overview

This project implements a pose classification system using:

• Base Model: MobileNetV2 (pre-trained on ImageNet)
• Architecture: Transfer learning with custom dense layers
• Training Strategy: Two-phase approach (frozen base + fine-tuning)
• Data Augmentation: Extensive augmentation for robust training

Results

• Test Accuracy: 91%
• Weighted F1-Score: 0.91
• Individual F1-Scores:
  • Standing: 0.93
  • Sitting: 0.88
  • Lying: 0.92

Dataset

The dataset contains:

• Standing: 421 images
• Sitting: 429 images
• Lying: 509 images
• Total: 1,359 images

Project Structure

human-pose-classifier/
├── notebooks/          # Jupyter notebooks
├── data/              # Dataset (lying, sitting, standing)
├── models/            # Trained models
├── results/          # Plots, metrics, and evaluation results
├── reports/          # Project documentation
├── venv/             # Virtual environment (created during setup)
├── requirements.txt  # Python dependencies
└── README.md         # This file

Setup

Prerequisites

• Python 3.8 or higher
• pip (Python package installer)

Installation

1. Clone the repository:

git clone https://github.com/sachin-prabhakar/human-pose-classifier.git
cd human-pose-classifier

2. Create a virtual environment:

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install required packages:

pip install -r requirements.txt

4. Register the virtual environment as a Jupyter kernel:

python -m ipykernel install --user --name=human-pose-classifier --display-name="Human Pose Classifier"

Running the Project

Option 1: Using Jupyter Notebook

# Activate virtual environment (if not already active)
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Start Jupyter Notebook
jupyter notebook notebooks/Human_Pose_Classification_Training.ipynb

Option 2: Using the registered kernel

1. Open Jupyter Notebook
2. Navigate to notebooks/Human_Pose_Classification_Training.ipynb
3. Go to Kernel → Change Kernel → Human Pose Classifier
4. Run the notebook cells

Deactivating the Virtual Environment

deactivate

Key Features

• Transfer Learning: MobileNetV2 pre-trained on ImageNet
• Data Augmentation: Rotation, flipping, brightness adjustment for robust training
• Class Balancing: Automatic weight computation for imbalanced dataset
• Two-Phase Training: Frozen base model + fine-tuning approach
• Comprehensive Evaluation: Confusion matrices, F1-scores, and accuracy metrics
• Real-world Testing: External image validation

Technologies Used

• TensorFlow/Keras: Deep learning framework
• MobileNetV2: Efficient CNN architecture for mobile deployment
• Scikit-learn: Model evaluation and metrics
• Pandas/NumPy: Data manipulation and numerical computing
• Matplotlib/Seaborn: Data visualization and plotting
• PIL (Pillow): Image processing and manipulation

Troubleshooting

Common Issues

1. ModuleNotFoundError for numpy, pandas, etc.

• Solution: Make sure you've activated the virtual environment and installed dependencies:

  source venv/bin/activate
  pip install -r requirements.txt

2. Jupyter kernel not found

• Solution: Register the virtual environment as a Jupyter kernel:

  source venv/bin/activate
  python -m ipykernel install --user --name=human-pose-classifier --display-name="Human Pose Classifier"

3. Permission denied errors on macOS

• Solution: Use virtual environment instead of system Python:

  python3 -m venv venv
  source venv/bin/activate

4. TensorFlow installation issues

• Solution: Ensure you have Python 3.8+ and install TensorFlow in the virtual environment:

  pip install tensorflow>=2.18.0

Getting Help

If you encounter issues not covered here:

1. Check that all dependencies are installed: pip list
2. Verify virtual environment is activated: which python
3. Ensure Jupyter is using the correct kernel
4. Check Python version compatibility: python --version