Lihong Wang

Research & Projects

Agentic multimodal reasoning

ViRC

An agentic multimodal model that improves visual interleaved mathematical reasoning through reason chunking and progressive training.

High-quality multimodal datasets

Human-verified aerial vision-language navigation datasets curated with ViRC-7B and 150 hours of manual verification.

Differentiable 3D rendering

A differentiable curve rendering framework for generating view-consistent 3D sketches under multiple forms of supervision.