mlsanityData to 14 Oct 2026

Unveiling Visual Perception in Language Models: An Attention Head Analysis Approach

arXiv 2412.18108 ↗First posted 24 Dec 2024cs.CVcited by 35

On the map: Multimodal LLMs / vision-language models → MLLM training and visual instruction tuning

Abstract (arXiv)

Recent advancements in Multimodal Large Language Models (MLLMs) have demonstrated remarkable progress in visual understanding. This impressive leap raises a compelling question: how can language models, initially trained solely on linguistic data, effectively interpret and process visual content? This paper aims to address this question with systematic investigation across 4 model families and 4 model scales, uncovering a unique class of attention heads that focus specifically on visual content. Our analysis reveals a strong correlation between the behavior of these attention heads, the distribution of attention weights, and their concentration on visual tokens within the input. These findings enhance our understanding of how LLMs adapt to multimodal tasks, demonstrating their potential to bridge the gap between textual and visual understanding. This work paves the way for the development of AI systems capable of engaging with diverse modalities.

Key facts. Cited by 35 papers in the map (data through 14 Oct 2026). Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 43 of its 61 references that are in the map. First posted on arXiv 24 Dec 2024.

Builds on

43 of 61 references.

Closest ancestors · 6

Same direction, newest first

  1. 2024Law of Vision Representation in MLLMs29 Aug
  2. 2023From CLIP to DINO: Visual Encoders Shout in Multi-modal Large Language Models13 Oct
  3. 2023Improved Baselines with Visual Instruction Tuning5 Oct
  4. 2023SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehension30 Jul
  5. 2023MMBench: Is Your Multi-modal Model an All-around Player?12 Jul
  6. 2023InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning11 May

Other references · 37

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 14,132
  4. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  5. 2020End-to-End Object Detection with Transformerscited by 7,414
  6. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,137
  7. 2023Mistral 7Bcited by 3,164
  8. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,917
  9. 2023Evaluating Object Hallucination in Large Vision-Language Modelscited by 1,729
  10. 2023Video-LLaMA: An Instruction-tuned Audio-Visual Language Model for Video Understandingcited by 1,471
  11. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,165
  12. 2022In-context Learning and Induction Headscited by 964
25 more
  1. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 856
  2. 2020Quantifying Attention Flow in Transformerscited by 775
  3. 2023A Comprehensive Overview of Large Language Modelscited by 648
  4. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
  5. 2023PMC-VQA: Visual Instruction Tuning for Medical Visual Question Answeringcited by 338
  6. 2023PandaGPT: One Model To Instruction-Follow Them Allcited by 334
  7. 2021Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformerscited by 320
  8. 2023Q-Bench: A Benchmark for General-Purpose Foundation Models on Low-level Visioncited by 249
  9. 2024Retrieval Head Mechanistically Explains Long-Context Factualitycited by 232
  10. 2023Video Understanding with Large Language Models: A Surveycited by 206
  11. 2023X-LLM: Bootstrapping Advanced Large Language Models by Treating Multi-Modalities as Foreign Languagescited by 128
  12. 2023DetGPT: Detect What You Need via Reasoningcited by 112
  13. 2024Information Flow Routes: Automatically Interpreting Language Models at Scalecited by 95
  14. 2024Attention Heads of Large Language Models: A Surveycited by 67
  15. 2021Procedure Planning in Instructional Videos via Contextual Modeling and Model-based Policy Learningcited by 53
  16. 2024Internal Consistency and Self-Feedback in Large Language Models: A Surveycited by 50
  17. 2020Point and Ask: Incorporating Pointing into Visual Question Answeringcited by 39
  18. 2023Visual Question Answering: A Survey on Techniques and Common Trends in Recent Literaturecited by 21
  19. 2024OSCaR: Object State Captioning and State Change Representationcited by 16
  20. 2024EAGLE: Egocentric AGgregated Language-video Enginecited by 16
  21. 2024Correcting Negative Bias in Large Language Models through Negative Attention Score Alignmentcited by 12
  22. 2024Analyzing The Language of Visual Tokenscited by 12
  23. 2023MISAR: A Multimodal Instructional System with Augmented Realitycited by 10
  24. 2024Investigating Neuron Ablation in Attention Heads: The Case for Peak Activation Centeringcited by 4
  25. 2019Learning from Interventions using Hierarchical Policies for Safe Learningcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 15; 2026: 20.

35 papers.

Sort:

By area:

  1. 2026Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Modelscited by 27
  2. 2025Why Reasoning Matters? A Survey of Advancements in Multimodal Reasoning (v1)cited by 23
  3. 2025Understanding Language Prior of LVLMs by Contrasting Chain-of-Embeddingcited by 14
  4. 2025Constructive Distortion: Improving MLLMs with Attention-Guided Image Warpingcited by 13
  5. 2025Caption Anything in Video: Fine-grained Object-centric Captioning via Spatiotemporal Multimodal Promptingcited by 10
  6. 2025How Do Large Vision-Language Models See Text in Image? Unveiling the Distinctive Role of OCR Headscited by 9
  7. 2025Multi-speaker Attention Alignment for Multimodal Social Interactioncited by 9
  8. 2026How Vision Becomes Language: A Layer-wise Information-Theoretic Analysis of Multimodal Reasoningcited by 8
  9. 2026Detecting Misbehaviors of Large Vision-Language Models by Evidential Uncertainty Quantificationcited by 6
  10. 2025MMPerspective: Do MLLMs Understand Perspective? A Comprehensive Benchmark for Perspective Perception, Reasoning, and Robustnesscited by 5
  11. 2025Seeing Beyond Words: Self-Supervised Visual Learning for Multimodal Large Language Modelscited by 5
  12. 2025Reallocating Attention Across Layers to Reduce Multimodal Hallucinationcited by 4
23 more
  1. 2026Steering and Rectifying Latent Representation Manifolds in Frozen Multi-modal LLMs for Video Anomaly Detectioncited by 4
  2. 2026Mechanistic Insights into Functional Sparsity in Multimodal LLMs via CoRe Headscited by 4
  3. 2025Investigating and Enhancing Vision-Audio Capability in Omnimodal Large Language Modelscited by 3
  4. 2025CAI: Caption-Sensitive Attention Intervention for Mitigating Object Hallucination in Large Vision-Language Modelscited by 3
  5. 2025When to Think and When to Look: Uncertainty-Guided Lookbackcited by 3
  6. 2026ViKey: Enhancing Temporal Understanding in Videos via Visual Promptingcited by 3
  7. 2025Video Parallel Scaling: Aggregating Diverse Frame Subsets for VideoLLMscited by 2
  8. 2025PATCH: Mitigating PII Leakage in Language Models with Privacy-Aware Targeted Circuit PatcHingcited by 2
  9. 2025Investigating The Functional Roles of Attention Heads in Vision Language Models: Evidence for Reasoning Modulescited by 2
  10. 2026Interpretable Motion-Attentive Maps: Spatio-Temporally Localizing Concepts in Video Diffusion Transformerscited by 1
  11. 2026Look Twice: Training-Free Evidence Highlighting for Knowledge-based Visual Question Answeringcited by 1
  12. 2026Head-wise Modality Specialization within MLLMs for Robust Fake News Detection under Missing Modalitycited by 1
  13. 2026TraversalBench: Challenging Paths to Follow for Vision Language Modelscited by 1
  14. 2026MLLMs Know When Before Speaking: Revealing and Recovering Temporal Grounding via Attention Cuescited by 1
  15. 2026Can Retrieval Heads See Images? Multimodal Retrieval Heads in Long-Context Vision-Language Modelscited by 1
  16. 2026AdaCount: Training-Free Similarity-Guided Spatial and Feature Adaptation for Zero-Shot Object Countingcited by 1
  17. 2026When Structured Sparse Autoencoders Learn Consistent Concepts Across Modalitiescited by 1
  18. 2026Knowing Isn't Always Saying: When Do Spatial Encodings Reach Answers in Vision-Language Models?cited by 1
  19. 2026System-Mediated Attention Imbalances Make Vision-Language Models Say Yescited by 0
  20. 2026Cognitive Alignment At No Cost: Inducing Human Attention Biases For Interpretable Vision Transformerscited by 0
  21. 2026CAST: Mitigating Object Hallucination in Large Vision-Language Models via Caption-Guided Visual Attention Steeringcited by 0
  22. 2026TextGaze: Prompting Gaze Target Estimation with Textual Scene Cuescited by 0
  23. 2026What Do Visual-Head Discovery Scores Actually Find? Attention Source, Necessity, and Random-Control Pitfalls in Vision-Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Perceiving Beyond Language Priors: Enhancing Visual Comprehension and Attention in Multimodal Models
  2. 2025Visual Representations inside the Language Model
  3. 2025Seeing is Understanding: Unlocking Causal Attention into Modality-Mutual Attention for Multimodal LLMs
  4. 2026Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention
  5. 2024A Comprehensive Survey and Guide to Multimodal Large Language Models in Vision-Language Tasks
  6. 2026Visual Attention Faithfulness in Vision-Language Models is Heterogeneous
  7. 2026Learning to See What You Need: Gaze Attention for Multimodal Large Language Models
  8. 2025Head Pursuit: Probing Attention Specialization in Multimodal Transformers