Figureure 2 · : Overview of the EADPFig. 2: Overview of the EADP. EADP acts as a plug-and-play module compressing N visual tokens into a highly informative subset of K tokens for the downstream LLM. Stage 1: An entropy-guided denoising mechanism filters out high-entropy textual noise to get the dense guidance score $S ^ { D }$ . This is fused with the global EOS score $S ^ { G }$ to yield a robust instruction relevance score $S ^ { I }$ . Stage 2: After refining $S ^ { I }$ via gaussian smoothing and score polarization to prevent feature fragmentation, a submodular maximization objective uses inter-token visual similarity to select the final subset Y.这张图概括 Combating Textual Noise and Redundancy 的整体方法流程。阅读时先看模块之间传递的训练信号，再看作者如何把目标拆成可优化的子问题。(a) (b) (c) Fig(a) (b) (c) Fig. 1: (a) illustrates a limitation of global guidance: it tends to attend to background regions. (b) highlights the dispersion phenomenon caused by textual noise. (c) reveals the issues of feature fragmentation and selection redundancy.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 Combating Textual Noise and Redundancy 的方法或实验，请结合正文精读段落一起看。
