Detached Skip-Links and $R$-Probe: Decoupling Feature Aggregation from Gradient Propagation for MLLM OCR 文章

ArXiv CS.CV2026-05-27NEWSen作者: Ziye Yuan, Ruchang Yao, Chengxin Zheng, Yusheng Zhao, Daxiang Dong, Ming Zhang
大语言模型
查看原文 →
关系图谱

Detached Skip-Links and $R$-Probe: Decoupling Feature Aggregation from Gradient Propagation for MLLM OCR · 相关技术

相关技术

Large language modelLLMMultimodal Large Language Models (MLLMs)ODEPLAlanguage model通用表达式语言多模态divide-and-conquer partitioningSEMReferring expression comprehension (REC)REUSEPromiseParts-of-Speech (POS) tagsPROBEOCRNarrative Abstraction BenchmarkMITMECLMMHISForEffort Metric AttentionARGANN