README.md

metadata

title: Yoga Pose Classifier
emoji: 🐢
colorFrom: blue
colorTo: yellow
sdk: docker
pinned: false
short_description: Add end-to-end yoga pose classification system with DenseNet

Yoga Pose Classification – Deep Learning Web Application

Overview

This project presents an end-to-end Yoga Pose Classification system that integrates Deep Learning (DenseNet121) with a Flask-based web application. Users can upload an image of a yoga pose and obtain the predicted pose along with confidence scores.

The project follows industry best practices, including modular code structure, clean separation of concerns, and deployment-ready architecture. It is suitable for portfolio presentation, internship submissions, and real-world AI application demos.

Key Skills and Technologies

• Deep Learning / Transfer Learning: DenseNet121, TensorFlow, Keras
• Computer Vision: Image preprocessing, augmentation, classification
• Web Development: Flask, Jinja templating, HTML/CSS
• Data Handling & Analysis: NumPy, Pandas, visualization
• Deployment Readiness: Modular structure, model serialization, upload handling

This highlights transferable skills relevant to AI, ML, and full-stack roles.

Dataset Information

• Dataset Name: Yoga Pose Classification Dataset
• Source: Kaggle
• Link: https://www.kaggle.com/datasets/ujjwalchowdhury/yoga-pose-classification
• Structure: Class-wise folders containing labeled images of yoga poses

---

Model Architecture

• Architecture: DenseNet121 (transfer learning)
• Pre-trained on: ImageNet
• Input Size: 224 × 224 × 3
• Loss Function: Categorical Crossentropy
• Optimizer: Adam (Learning Rate: 1e-4)
• Evaluation Metrics: Accuracy, Precision, Recall, F1-score

The DenseNet base layers are frozen and a custom classification head is trained on the yoga pose dataset.

---

Project Structure

yoga-pose-classifier/
│
├── app.py                     # Main Flask application
├── requirements.txt           # Project dependencies
├── notebook/                  # Jupyter notebook containing 
│   └── yoga_pose_classification.ipynb
│
├── model/
│   └── model_dense121.keras   # Trained DenseNet121 model
│
├── utils/
│   ├── allowed_file.py        # File extension validation
│   └── upload_file.py         # Upload redirection helper
│
├── static/
│   ├── css/
│   │   └── style.css     # styling
│   └── uploads/          # Uploaded images
│
├── templates/
│   └── index.html             # Main UI template
└── README.md             

Application Workflow

1. User uploads an image via the web interface.
2. File is validated using allowed_file().
3. Image is saved to static/uploads/.
4. Image is preprocessed for DenseNet121 input.
5. Model predicts pose and confidence.
6. Result displayed on UI.

Utility Modules

• allowed_file.py: Ensures only supported image formats are accepted.
• upload_file.py: Handles clean routing for uploaded images.

Modular utilities improve code readability and maintainability.

Installation & Setup

1. Clone the Repository

git clone https://github.com/batoolarifa/yoga-pose-classifier
cd yoga-pose-classifier

2. Install Dependencies

pip install -r requirements.txt

3. Run the Application

python app.py

Access the app at: http://localhost:8080

Model Inference Example

• Upload a yoga pose image (JPG / PNG).

• Model predicts one of the following poses:

  • Downdog
  • Goddess
  • Plank
  • Tree
  • Warrior2

• Confidence score displayed alongside prediction.

Deployment

• Ready for deployment on platforms such as Hugging Face Spaces, Render, AWS EC2, or Docker environments.
• Model loaded once at startup for efficient inference.
• Supports easy class extension and integration into larger systems.

Industry Relevance & Value

This project demonstrates:

• End-to-end ML application development
• Deep learning expertise
• Full-stack AI system implementation
• Reproducible and scalable code practices
• Skills aligned with AI/ML, Computer Vision, and Software Engineering roles

This section highlights the project’s alignment with industry standards and professional portfolios.

Future Improvements

• Real-time webcam inference
• Pose correction feedback and tips
• REST API conversion (FastAPI)
• Enhanced UI accessibility and responsiveness

👤 Author

Syeda Arifa Batool
SE @ Karachi University | AI & ML Practitioner | Applying technology to create real-world value 📈

🔗 Connect with Me

• LinkedIn: https://www.linkedin.com/in/arifa-batool/
• Kaggle: https://www.linkedin.com/in/arifa-batool/
• Email: thearifabatool@gmail.com

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference