Towards Interpreting Visual Information Processing in Vision-Language Models
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:ID67LVCXrecord.jsonopen to challenge →
read the original abstract
Vision-Language Models (VLMs) are powerful tools for processing and understanding text and images. We study the processing of visual tokens in the language model component of LLaVA, a prominent VLM. Our approach focuses on analyzing the localization of object information, the evolution of visual token representations across layers, and the mechanism of integrating visual information for predictions. Through ablation studies, we demonstrated that object identification accuracy drops by over 70\% when object-specific tokens are removed. We observed that visual token representations become increasingly interpretable in the vocabulary space across layers, suggesting an alignment with textual tokens corresponding to image content. Finally, we found that the model extracts object information from these refined representations at the last token position for prediction, mirroring the process in text-only language models for factual association tasks. These findings provide crucial insights into how VLMs process and integrate visual information, bridging the gap between our understanding of language and vision models, and paving the way for more interpretable and controllable multimodal systems.
This paper has not been read by Pith yet.
Forward citations
Cited by 16 Pith papers
-
MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs
MVI-Bench supplies the first taxonomy and dataset focused on misleading visual inputs to measure LVLM robustness, with tests on 18 models revealing clear weaknesses.
-
Vision-Default, Prior-Override: Causal Mechanisms of Perception-Knowledge Conflict in Vision-Language Models
VLMs default to visual grounding but a sparse circuit of 2.5-4.8% attention heads in later layers mediates prior-knowledge overrides, identified causally via patching and ablation across three model families.
-
From Senses to Decisions: The Information Flow of Auditory and Visual Perception in Multimodal LLMs
AVLLMs route audio-visual information sequentially in video tasks and via parallel streams for interleaved items, allowing early token discard with little performance loss across models and scales.
-
ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety
ProjLens shows that backdoor parameters in MLLMs are encoded in low-rank subspaces of the projector and that embeddings shift toward the target direction with magnitude linear in input norm, activating only on poisone...
-
V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models
V-SEAM combines concept-level visual semantic editing with attention head modulation to identify positive and negative contributors across object, attribute, and relationship levels, then uses this to improve VLM perf...
-
Analysis-by-Proxy: Localization Signals in VLMs Operating as Condition Encoders
A lightweight Q-Former proxy trained on VLM hidden states reveals that localization signals peak in input-dependent intermediate layers, not the final layers used by standard editing pipelines.
-
When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models
Decoder-based VLMs hallucinate due to geometric over-alignment of visual embeddings with the text manifold in a universal dataset-agnostic subspace, mitigated by projecting out the linguistic bias.
-
When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models
Decoder-based VLMs over-align visual features to a universal text subspace, injecting linguistic bias; projecting out its top principal components reduces hallucinations on POPE, CHAIR, AMBER and improves long-form ca...
-
When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models
Decoder-based VLMs hallucinate because visual embeddings are over-aligned to a text manifold; projecting out the top principal components of a universal linguistic subspace reduces this bias and improves benchmark per...
-
When Language Overwrites Vision: Over-Alignment and Geometric Debiasing in Vision-Language Models
Decoder-based VLMs over-align visual embeddings to text manifold causing linguistic bias in top PCs of a universal text subspace; projecting out this subspace reduces hallucinations on POPE/CHAIR/AMBER and improves CLAIR.
-
CAST: Mitigating Object Hallucination in Large Vision-Language Models via Caption-Guided Visual Attention Steering
CAST reduces object hallucination in LVLMs by 6.03% on average across five models and five benchmarks by identifying caption-sensitive attention heads and applying optimized steering directions to their outputs, with ...
-
STEAR: Layer-Aware Spatiotemporal Evidence Intervention for Hallucination Mitigation in Video Large Language Models
STEAR reduces spatial and temporal hallucinations in Video-LLMs via layer-aware evidence intervention from middle decoder layers in a single-encode pass.
-
VLMs Need Words: Vision Language Models Ignore Visual Detail In Favor of Semantic Anchors
VLMs bypass visual comparison by recovering semantic labels for nameable entities and hallucinate on unnamable ones, as shown by performance gaps and Logit Lens analysis.
-
Understanding Counting Mechanisms in Large Language and Vision-Language Models
LLMs and LVLMs encode latent positional count information in individual tokens or visual features, with an internal counter mechanism that updates per item and emerges progressively across layers, relying on structura...
-
The Hidden Evolution of Disguised Visual Context inside the VLM
Visual tokens enter VLMs as raw signals and are reshaped differently by in-context versus layer-injection paradigms, each capturing distinct frequency characteristics that drive task performance.
-
From Heads to Neurons: Causal Attribution and Steering in Multi-Task Vision-Language Models
HONES ranks feed-forward neurons by their causal contributions from task-relevant attention heads and uses lightweight scaling to steer performance on multiple vision-language tasks.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.