#spatial grounding

Can Vision-Language Models Assess Proxemic Risk from Egocentric⭐10

Current VLMs show limits in egocentric proxemic risk assessment, though targeted prompts and fine-tuning boost high-danger recall in models like Qwen-VL.