IDD-X

IDD-X is a large-scale, dual-view driving video dataset designed to advance understanding of how road conditions and surrounding traffic influence the ego vehicle’s driving behavior, particularly in dense, heterogeneous, and unstructured traffic scenarios common in developing countries. Unlike existing datasets focused on structured environments, IDD-X captures the complexity of real-world traffic through 3634 annotated driving scenarios across 1140 untrimmed videos. It features 697K bounding boxes, 9K object tracks, and annotations for 1–12 important road objects per scenario, spanning 10 object categories and 19 explanation label categories. The dataset includes both front and rear views, enabling a more complete and explainable analysis of driving decisions. To support this, custom-designed deep networks are introduced for localizing important objects and predicting per-object behavioral explanations, forming a foundation for safe and efficient navigation in intelligent vehicle systems.