Sitting Posture Analysis

This project analyzes human sitting posture from images and uses keypoint extraction and classification techniques to assess ergonomics. It leverages TensorFlow for deep learning, DVC for data versioning, and has several Jupyter Notebooks for interactive exploration.

Project Structure

• .dvc/
  Contains DVC configuration and remote settings (see .dvc/config).
• data/
  Holds original data and generated outputs (CSV files, processed images). Note that some files are ignored by .gitignore (e.g. /data.csv, /poses_images_out, /processed).
• dvclive/
  Contains subdirectories for different model experiments (e.g. adaboost/, nn/, xgb/). Each folder has its respective output plots and metrics.
• notebooks/
  Jupyter notebooks for interactive experiments:
  • classification.ipynb – performs pose classification using extracted landmarks.
  • keypoints_extraction.ipynb – demonstrates keypoint detection and CSV saving (see excerpts starting at line 32 and line 663).
• src/
  Python scripts and modules:
  • featurize.py – extracts features from the original images for further analysis.
  • prepare.py – processes the extracted data and writes train, validation, and test CSVs (refer to src/prepare.py around line 106).
  • data.py and evaluate.py – contain data handling and evaluation routines.
• config.lua
  A Lua configuration file which may be used to set project parameters.
• requirements.txt
  Lists all Python dependencies (for example, pillow, polars, psutil, etc.). Check the file for dependency details (see excerpts starting at line 29, 191, and 289).

Setup and Installation

1. Clone the Repository:

   git clone <repository-url>
   cd sitting-posture

2. Environment Setup:

   Create a virtual environment (e.g. using venv or conda) and install dependencies:

   python -m venv .venv
   source .venv/bin/activate      # On Windows use .venv\Scripts\activate
   pip install -r requirements.txt

3. DVC Setup:

   Ensure DVC is installed. The project uses DVC to manage large datasets. Check dvc.yaml and dvc.lock for pipeline stages and dependencies.

   To pull the data from the remote, run:

   dvc pull

Running the Pipelines and Notebooks

• Data Featurization:

  Run the featurization stage as defined in dvc.lock:

  python src/featurize.py

• Prepare Data:

  After featurization, the prepare.py script writes CSVs into data/processed that are used for training and evaluation:

  python src/prepare.py

• Interactive Notebook Execution:

  Open notebooks such as classification.ipynb and keypoints_extraction.ipynb in Visual Studio Code or Jupyter to explore the extraction and classification workflows. These notebooks show how images are processed, CSVs are generated, and how the classification model is defined (see excerpt around line 250).

Model Training and Evaluation

The classification model uses TensorFlow:

• The input layer is defined with shape 39 for the landmark embedding.
• Dense layers with dropout are used for robustness.
• Model summary information is printed upon instantiation (refer to classification.ipynb starting at line 250).

Code Quality Tools

• Ruff (configured in .idea/ruff.xml) is set to run on save for style enforcement and linting.
• Inspection Profiles in the IntelliJ IDEA project configuration (.idea/inspectionProfiles/Project_Default.xml) list packages and modules which are monitored.

Contributing

Feel free to open issues or submit pull requests if you wish to improve the project. Make sure to:

• Follow the existing code style.
• Update the README if you add major functionality.
• Test your changes locally before committing.

References

• DVC Documentation
• TensorFlow Tutorials
• Project Structure Guidelines