mlsanityData to 7 Oct 2026

Semantic-Preserving Cross-Style Visual Reasoning for Robust Multi-Modal Understanding in Large Vision-Language Models

arXiv 2510.22838 ↗First posted 26 Oct 2025cs.CVcited by 0

On the map: LLM reasoning → Multimodal reasoning

also in 1 more
Abstract (arXiv)

The "style trap" poses a significant challenge for Large Vision-Language Models (LVLMs), hindering robust semantic understanding across diverse visual styles, especially in in-context learning (ICL). Existing methods often fail to effectively decouple style from content, hindering generalization. To address this, we propose the Semantic-Preserving Cross-Style Visual Reasoner (SP-CSVR), a novel framework for stable semantic understanding and adaptive cross-style visual reasoning. SP-CSVR integrates a Cross-Style Feature Encoder (CSFE) for style-content disentanglement, a Semantic-Aligned In-Context Decoder (SAICD) for efficient few-shot style adaptation, and an Adaptive Semantic Consistency Module (ASCM) employing multi-task contrastive learning to enforce cross-style semantic invariance. Extensive experiments on a challenging multi-style dataset demonstrate SP-CSVR's state-of-the-art performance across visual captioning, visual question answering, and in-context style adaptation. Comprehensive evaluations, including ablation studies and generalization analysis, confirm SP-CSVR's efficacy in enhancing robustness, generalization, and efficiency across diverse visual styles.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 14 of its 25 references that are in the map. First posted on arXiv 26 Oct 2025.

Builds on

14 of 25 references.

References · 14

  1. 2020Making Pre-trained Language Models Better Few-shot Learnerscited by 1,255
  2. 2021MetaICL: Learning to Learn In Contextcited by 450
  3. 2021Visually Grounded Reasoning across Languages and Culturescited by 188
  4. 2021OpenPrompt: An Open-source Framework for Prompt-learningcited by 148
  5. 2024Visual In-Context Learning for Large Vision-Language Modelscited by 124
  6. 2021A Good Prompt Is Worth Millions of Parameters: Low-resource Prompt-based Learning for Vision-Language Modelscited by 101
  7. 2022Thinking about GPT-3 In-Context Learning for Biomedical IE? Think Againcited by 83
  8. 2021What's in the Box? A Preliminary Analysis of Undesirable Content in the Common Crawl Corpuscited by 79
  9. 2021Towards Interpreting and Mitigating Shortcut Learning Behavior of NLU Modelscited by 78
  10. 2020Measuring Social Biases in Grounded Vision and Language Embeddingscited by 61
  11. 2021Temporal Adaptation of BERT and Performance on Downstream Document Classification: Insights from Social Mediacited by 51
  12. 2022Vision-Language Pre-Training for Multimodal Aspect-Based Sentiment Analysiscited by 21
2 more
  1. 2024Enhanced Visual SLAM for Collision-free Driving with Lightweight Autonomous Carscited by 6
  2. 2024Detect an Object At Once without Fine-tuningcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2024Enhancing Advanced Visual Reasoning Ability of Large Language Models
  2. 2025Interleaved Latent Visual Reasoning with Selective Perceptual Modeling
  3. 2024Dynamic Adapter with Semantics Disentangling for Cross-lingual Cross-modal Retrieval
  4. 2025Context-Aware Multi-Turn Visual-Textual Reasoning in LVLMs via Dynamic Memory and Adaptive Visual Guidance
  5. 2024Unraveling Cross-Modality Knowledge Conflicts in Large Vision-Language Models
  6. 2026CoRe: A Comprehensive Framework for Cross-Image Comparative Reasoning in Vision-Language Models
  7. 2025CF-VLM:CounterFactual Vision-Language Fine-tuning
  8. 2026CoVFT: Context-aware Visual Fine-tuning for Multimodal Large Language Models