Data Science Project UC3M - Face Detection & Head Pose Estimation

Overview

This project benchmarks face detection models and explores head pose estimation as part of a Data Science course at UC3M. The goal is to evaluate various face detection algorithms across different metrics including accuracy, precision, and inference time.

Project Structure

• benchmark/ - Face detection benchmark framework with model implementations and evaluation scripts
• docs/ - Project documentation, papers analysis, and state-of-the-art research
• pipeline/ - Data pipeline utilities for dataset processing
• report/ - LaTeX source files for the final project report
• visualization/ - Streamlit-based visualization application for results

Features

• Comprehensive benchmarking of multiple face detection models:
  • YOLO variants (YOLOv5, YOLOv8, YOLOv10, YOLOv11)
  • Classical approaches (Haar Cascade, SSD)
  • Modern detectors (FaceBoxes, LFFD, DSFD, RetinaFace, MTCNN, BlazeFace, SCRFD)
  • Lightweight models (Ultra-Light-Fast-Generic-Face-Detector)
• Evaluation on WIDER FACE dataset
• Performance metrics: accuracy, precision, recall, inference time
• Interactive visualization dashboard

Installation

Prerequisites

• Python 3.8+
• CPU-only environment (no GPU required)

Setup

1. Clone the repository:

git clone https://github.com/andres-nav/data-science-project-uc3m.git
cd data-science-project-uc3m

2. Install dependencies:

# For benchmark framework
cd benchmark
pip install -r requirements.txt

# For visualization app
cd ../visualization
pip install -r requirements.txt

3. Initialize submodules:

git submodule update --init --recursive

Usage

Running Benchmarks

cd benchmark
python main.py

Visualization Dashboard

cd visualization
streamlit run app.py

Results

Detailed benchmark results, model comparisons, and analysis can be found in the final report PDF located in the report/ directory.