https://github.com/theosorus/formulatracker

Training a YOLO model on a custom Formula 1 dataset to detect cars based on their team.
https://github.com/theosorus/formulatracker

computer-vision deep-learning formula1 tracking yolo

Last synced: about 1 month ago
JSON representation

Training a YOLO model on a custom Formula 1 dataset to detect cars based on their team.

• Host: GitHub
• URL: https://github.com/theosorus/formulatracker
• Owner: theosorus
• License: agpl-3.0
• Created: 2025-03-20T08:32:26.000Z (7 months ago)
• Default Branch: main
• Last Pushed: 2025-08-22T18:25:31.000Z (about 2 months ago)
• Last Synced: 2025-08-22T21:02:15.713Z (about 2 months ago)
• Topics: computer-vision, deep-learning, formula1, tracking, yolo
• Language: Python
• Homepage:
• Size: 3.37 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 🏎️ FormulaTracker — Detect F1 cars by team with YOLO

> The goal of this project is to train a **YOLO model** on a **custom dataset** to detect Formula 1 cars and classify them by team in video footage.



  





  Dataset I built on Kaggle •

  Ultralytics YOLO docs •



---

# 📊 Dataset

- **Source:** curated from a full Grand Prix broadcast. Non-relevant segments were trimmed out.

- **Annotation tool:** [labelImg](https://github.com/HumanSignal/labelImg)

- **Split:** `train = 442` images, `val = 111` images

- **Classes (10):**

  

| Team              |

|-------------------|

| Alfa Romeo Racing |

| Ferrari           |

| Haas              |

| McLaren           |

| Mercedes          |

| Racing Point      |

| RedBull           |

| Renault           |

| Toro Rosso        |

| Williams          |

# 🏋️ Train the model

| Hyperparameters | value      |

|-----------------|------------|

| task            | detect     |

| mode            | train      |

| model           | yolo11l.pt |

| epochs          | 200        |

| batch           | 16         |

| imgsz           | 640        |

### Results

  

### Confusion matrix

  

### Sample Predictions (validation batch)

  

# 🚀 Roadmap / Future ideas

- 🚥 Real-time speed estimation: approximate car speeds using multi-frame tracking + homography.

- 📺 On-screen overlay: draw team labels on live or recorded video streams.

- 🧩 Tracking: integrate ByteTrack/BoT-SORT for consistent track IDs across frames.

- 🏁 More seasons: expand dataset with multiple races and lighting/weather conditions.