Task. Critical Object Grounding: Given an accident video clip, our goal is to output bounding boxes associated with human driver gaze fixations, which are considered critical objects. Specifically, we encourage participants to use the DADA-2000 dataset (Train) as a benchmark and incorporate multimodal large models（MLLM）for network modeling.