Mengdi Jia

I am currently a mechatronics engineer based in Beijing. Previously, I received my Master's degree from Anhui Agricultural University, advised by Prof. Chengmao Cao, and my Bachelor's degree from College of Modern Science And Technology Hebei Agricultural University. During my research internship at Tsinghua University, I contributed to projects including OmniSpatial, working closely with Zekun Qi.

My primary focus is on Embodied Intelligence, and Multimodal Large Language Models.

Email  /  Scholar  /  GitHub  /  CV

Publications

* indicates equal contribution

We present OmniSpatial, a benchmark for evaluating spatial reasoning abilities in vision-language models. It covers four major categories and 50+ subtypes, totaling over 1.5K QA pairs. Experiments reveal significant limitations in existing models and suggest directions for improvement.

[arXiv] [Project Page] [Code] [Huggingface]