SynWTS: Multi-View Digital-Twin Dataset for Synthetic-to-Real Traffic Safety Understanding
Published in 2026 European Conference on Computer Vision - 10th AI City Challenge Workshop, 2026
Recommended citation: Dhanishtha Patil, Ridham Kachhadiya, Andrew Vattuone, Haoquan Liang, Jiajun Li, Yuxing Wang, Zheng Tang, Ashutosh Kumar, Quan Kong and David C. Anastasiu. "SynWTS: Multi-View Digital-Twin Dataset for Synthetic-to-Real Traffic Safety Understanding". Proceedings of European Conference on Computer Vision Workshops (ECCVW 2026). 2026. https://openreview.net/forum?id=BbaZpvqrel
[Paper]
Abstract
SynWTS is a multi-view synthetic digital twin of the real-world Woven Traffic Safety dataset for synthetic-to-real traffic safety understanding. It provides 249 paired scenarios and 846 temporally synchronized videos rendered from overhead and ego-vehicle perspectives, with frame-level bounding boxes, dense textual descriptions, visual question-answering pairs, and staged accident phases. By pairing each synthetic video with a corresponding real recording, SynWTS enables direct measurement of sim-to-real transfer and analysis of how visual realism affects downstream traffic-safety reasoning.
BibTeX
@InProceedings{Patil26SynWTS,
author = {Dhanishtha Patil and Ridham Kachhadiya and Andrew Vattuone and Haoquan Liang and Jiajun Li and Yuxing Wang and Zheng Tang and Ashutosh Kumar and Quan Kong and David C. Anastasiu},
title = {{SynWTS}: {M}ulti-view digital-twin dataset for synthetic-to-real traffic safety understanding},
booktitle = {Proc. ECCV Workshops},
address = {Malm{"o}, Sweden},
year = {2026}
}