arXiv:2509. 22415v3 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) have achieved strong vision-language performance, yet their token-level visual evidence remains difficult to inspect.
TextVBench is a new benchmark and evaluation framework for assessing how well video generation models render and preserve text. It introduces two tracks—TextVBench‑T2V for prompt‑specified text creation and maintenance, and TextVBench‑Ref2V for preserving text from reference images—using a multi‑target, multi‑instance spatiotemporal correspondence approach. The framework employs target‑conditioned candidate selection, supports multiple OCR regions per target, and separates spatial and temporal fidelity, summarizing results with a length‑weighted harmonic score and detailed diagnostics.
arXiv:2607. 03358v1 Announce Type: cross Abstract: We study how visual information is routed in vision-language models (VLMs).
arXiv:2610.19407v1 Announce Type: new Abstract: Vision-language models must reconcile conflicting text and images. Prior work has characterized the resulting behavior in terms of modality biases--e.g...
The paper introduces Visual Retrieval Heads (VRHs), a small fraction of attention heads in vision‑language models that are causally responsible for grounding text descriptions to image regions. By recasting head‑scoring methods and evaluating across eleven VLMs and five benchmarks, the authors show that masking the top 20 VRHs can drop grounding accuracy by up to 80 percentage points, while random masking has little effect. VRHs generalize across various visual reference tasks, preserve output format while corrupting localization, and transfer causally across models sharing an LLM backbone.
As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanis...
The paper introduces a controlled anchor–stress protocol to analyze compositional failures in text‑to‑image diffusion models. It presents a text‑only Compositional Stress Index (CSI) that distinguishes common from rare compositions across SD1.5, SDXL, and SD3, and conducts a six‑prompt localization study showing how different denoiser interventions affect outcomes such as residual minimization and color hit rate. The study reveals that diagnostic signals appear before denoising, but the intervention that exposes risk is not always the one that improves generation, highlighting modality‑ and architecture‑dependent responses.
The paper demonstrates that vision‑language models (VLMs) possess a small set of attention heads, called Visual Retrieval Heads (VRHs), that are causally responsible for linking text prompts to specific image regions. By adapting head‑scoring techniques from language models, the authors identify VRHs as the heads whose attention from output prediction tokens, summed over the ground‑truth referent region, most reliably indicates causal grounding. Experiments across eleven VLMs and five referring‑expression benchmarks show that masking the top 20 VRHs can drop grounding accuracy by up to 80 percentage points, while random masking has little effect, and that VRHs generalize across diverse visual tasks and transfer across models sharing an LLM backbone.
arXiv:2609.36598v1 Announce Type: new Abstract: A video can exhibit convincing motion and photorealism yet fail immediately when visual text collapses. Unlike generic scene content, visual text is un...
arXiv:2602. 00344v2 Announce Type: replace-cross Abstract: While Retrieval-Augmented Generation (RAG) is one of the dominant paradigms for enhancing Large Vision-Language Models (LVLMs) on knowledge-based VQA tasks, recent work attributes RAG failures to insufficient attention towards the retrieved context, proposing to reduce the attention allocated to image tokens.
arXiv:2609.38810v1 Announce Type: cross Abstract: As vision language models are increasingly deployed in clinical diagnosis, understanding how they internally resolve competing visual and textual sig...
arXiv:2609.06011v1 Announce Type: cross Abstract: Omni-modal large language models (OLLMs) jointly process vision, audio, and text, yet their modality bias under cross-modal conflict remains underexp...