Dataset for Paper "Metric Learning–Based Latent Space Construction for Edge-Case Detection in Multi-Agent Traffic Scenarios"

This dataset is a part of the paper published in IEEE Conference on Intelligent Vehicles 2026 (IV26), " Metric Learning–Based Latent Space Construction for Edge-Case Detection in Multi-Agent Traffic Scenarios." The provided datasets consist of preprocessed samples generated for metric learning. A data mining pipeline was applied to extract relevant traffic interactions and generate cinquplets that contain positive and negative examples for the metric learning task. These samples are used for training and evaluation of the proposed approach. The dataset is divided into two parts: data_metric_interaction contains the generated metric learning samples derived from the INTERACTION dataset. This dataset was used for pretraining the metric learning model and provides a large-scale set of multi-agent traffic interaction examples. data_metric_real contains the generated metric learning samples derived from real-world traffic data collected at an urban intersection in Ingolstadt, Germany, as well as from a reconstructed intersection at a test facility. This dataset was used for evaluation of the proposed approach on real-world traffic scenarios.