Liangcai Gao

We propose Uni-MuMER, full fine-tuning a VLM for handwritten math expression recognition via structured spatial reasoning (Tree-CoT), error-driven learning (EDL), and symbol counting (SC), achieving state-of-the-art results.