English · 中文 · Polski · Español · 日本語 · 한국어 · Русский · Français · Deutsch

Collision Detection AI

Collision detection for accelerometer data with a lightweight PyTorch model (1.3k parameters) built for real-time windowed inference.

Simple Setup

python -m venv .venv && source .venv/bin/activate
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu132
pip install -r requirements.txt

collision-detection all          # train (generates data, trains, evaluates)
collision-detection evaluate     # test the trained model on all splits
collision-detection visualize    # visualization of the trained model's tests

All commands produce artifacts in data/ (datasets) and artifacts/ (checkpoint, metrics, plots — including artifacts/plots/visualization.png).

For your own data, replace the CSVs in data/raw/ and run collision-detection preprocess before collision-detection train.

Citation

If this repository has helped you during your research, feel free to cite it:

APA Style

Brzustowicz, S. (2026). Collision-Detection-AI: Collision detection from accelerometer data using a lightweight PyTorch model (Version 1.0.0) [Source code]. https://github.com/sebastianbrzustowicz/Collision-Detection-AI

BibTeX

@software{brzustowicz_collision_detection_ai_2026,
  author = {Sebastian Brzustowicz},
  title = {Collision-Detection-AI: Collision detection from accelerometer data using a lightweight PyTorch model},
  url = {https://github.com/sebastianbrzustowicz/Collision-Detection-AI},
  version = {1.0.0},
  year = {2026}
}

Tip

You can also use the "Cite this repository" button in the sidebar to automatically copy these citations or download the raw metadata file.

License

CC BY-NC-ND 4.0 — see LICENSE.md.

Author

Sebastian Brzustowicz Se.Brzustowicz@gmail.com