Frontier research on Visual Intelligence: see further, build together.
We are the Visual Understanding Group within Xiaomi's Large Language Model (LLM) Applications Team (@XiaoMi · @xiaomi-research).
We study how AI systems perceive, understand, and interact with the visual world. Our research spans multimodal reasoning, document understanding, and GUI agents, with an emphasis on learning and evaluation in real-world settings.
| Project | Focus | Resources |
|---|---|---|
| Xiaomi-GUI | Multimodal GUI agents for real mobile environments, combining interface understanding, multi-step planning, and error recovery. Includes the RealMobile benchmark and evaluation tools. | Code · Report · Project Page |
| Xiaomi-OCR | Document parsing and OCR-centric understanding: structured Markdown with tables and formulas, key information extraction, and visual question answering. Includes inference pipelines and tools for agent integration. | Code · Model · Report |
We share research code, models, and evaluation resources to support reproducible research. We welcome bug reports, pull requests, benchmark contributions, and research collaborations. Start a conversation through an issue in the relevant repository.