Reproduce “Robust Anomaly Detection through Multi-Modal Autoencoder Fusion for Small Vehicle Damage Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sara Khan, Mehmed Yüksel, Frank Kirchner
Researchers developed a multi-modal autoencoder fusion system combining IMU and microphone data for robust, real-time detection of minor vehicle damage. This system achieved an ROC-AUC of 0.92, outperforming mono-modal baselines and more complex architectures, while maintaining efficiency suitable for edge deployment.
Key insights
- Multi-modal fusion of IMU and audio data, specifically with a pooling-based autoencoder (MAA3), significantly enhances minor vehicle damage detection accuracy over unimodal or simpler fusion methods.
- An autoencoder-based anomaly detection framework is effective for this task, reducing the reliance on extensive and costly labeled datasets of various damage types by learning only "normal" system behavior.
- The Log-Cosh loss function proved most robust for training autoencoders in this context, effectively handling outliers and contributing to higher performance compared to MSE or SSIM.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.