Video diffusion techniques have advanced significantly in recent years; however, they struggle to generate realistic imagery of car crashes due to the scarcity of accident events in most driving datasets. Improving traffic safety requires realistic and controllable accident simulations. To tackle the problem, we propose Ctrl-Crash, a controllable car crash video generation model that conditions on signals such as bounding boxes, crash types, and an initial image frame. Our approach enables counterfactual scenario generation where minor variations in input can lead to dramatically different crash outcomes. To support fine-grained control at inference time, we leverage classifier-free guidance with independently tunable scales for each conditioning signal. Ctrl-Crash achieves state-of-the-art performance across quantitative video quality metrics (e.g., FVD and JEDi) and qualitative measurements based on a human-evaluation of physical realism and video quality compared to prior diffusion-based methods.
These examples illustrate scenarios conditioned for multiple distinct crash types (which describe which actors are involved in the crash):
|
No Crash
|
Ego-Only crash
|
Ego-and-Vehicle crash
|
Vehicle-Only crash
|
Vehicle-and-Vehicle crash
|
Crash predicted by Ctrl-Crash using only the initial ground-truth frame and all bounding-box frames as input:
Crash predicted by Ctrl-Crash using the initial frame and the first 9 bounding-box frames as input (white frames indicate that the bounding-boxes were masked):
Generating crashes from the non-accident BDD100K dataset by conditioning on the initial frame and the first 9 bounding-box frames:
Other methods struggle to generate realistic crashes
@misc{gosselin2025ctrlcrashcontrollablediffusionrealistic,
title={Ctrl-Crash: Controllable Diffusion for Realistic Car Crashes},
author={Anthony Gosselin and Ge Ya Luo and Luis Lara and Florian Golemo and Derek Nowrouzezahrai and Liam Paull and Alexia Jolicoeur-Martineau and Christopher Pal},
year={2025},
eprint={2506.00227},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2506.00227},
}