Intersection-VideoQA

The CarlaGen Multi-modal Traffic Intersection Dataset is a synthetic large-scale multi-modal dataset for traffic scene understanding and autonomous driving perception research, generated using the official open-source CarlaGen pipeline based on CARLA simulator. It focuses on complex urban intersection scenarios, covering rich traffic participant interactions, traffic light states, and road topology information, providing multi-dimensional annotation support for both traditional perception tasks and emerging vision-language tasks. The dataset is split as follows: Towns 01 to 07 serve as the training set, and Town 10 serves as the validation set.