HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token

• URL: http://arxiv.org/abs/2603.05465v1
• Date: Thu, 05 Mar 2026 18:36:31 GMT
• Title: HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token
• Abstract summary: Hallucinations remain a persistent challenge for vision-language models (VLMs)<n>We investigate whether hallucination risk can be predicted before any token is generated by probing a model's internal representations in a single forward pass.
• Score: 4.2940165358265086
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Hallucinations remain a persistent challenge for vision-language models (VLMs), which often describe nonexistent objects or fabricate facts. Existing detection methods typically operate after text generation, making intervention both costly and untimely. We investigate whether hallucination risk can instead be predicted before any token is generated by probing a model's internal representations in a single forward pass. Across a diverse set of vision-language tasks and eight modern VLMs, including Llama-3.2-Vision, Gemma-3, Phi-4-VL, and Qwen2.5-VL, we examine three families of internal representations: (i) visual-only features without multimodal fusion, (ii) vision-token representations within the text decoder, and (iii) query-token representations that integrate visual and textual information before generation. Probes trained on these representations achieve strong hallucination-detection performance without decoding, reaching up to 0.93 AUROC on Gemma-3-12B, Phi-4-VL 5.6B, and Molmo 7B. Late query-token states are the most predictive for most models, while visual or mid-layer features dominate in a few architectures (e.g., ~0.79 AUROC for Qwen2.5-VL-7B using visual-only features). These results demonstrate that (1) hallucination risk is detectable pre-generation, (2) the most informative layer and modality vary across architectures, and (3) lightweight probes have the potential to enable early abstention, selective routing, and adaptive decoding to improve both safety and efficiency.

Related papers

• UniProbe: A Learnable Token-Level Hallucination Detector for Large VLMs using Multi-Structural Internal Representations [61.85916280854257]
  Large Vision-Language Models (LVLMs) achieve impressive visual reasoning and dialogue capabilities, yet frequently hallucinate content unsupported by the visual input.<n>We introduce textbfUniProbe, a lightweight, unified, learnable detector that models a frozen LVLM's heterogeneous computational trace from a single forward pass.
  arXiv  Detail & Related papers  (2026-08-11T12:01:59Z)
• Role-Break in Attention Heads: Understanding and Detecting Hallucinations in VLMs [49.743008490747094]
  Vision-Language Models (VLMs) remain prone to hallucinations, producing content that is inconsistent with or unsupported by the input image.<n>Under a unified head-level view, we find that hallucination-induced changes manifest as localized deviations from each head's faithful contextual behavior.<n> Detailed analysis reveals that these deviations are systematically organized across attention heads, contextual sources, and deviation directions, and that the resulting signal is linearly readable once head identity is preserved.
  arXiv  Detail & Related papers  (2026-07-31T13:33:51Z)
• Seeing or Knowing? Visual Context Sensitivity in Multimodal Large Language Models [14.5102059776721]
  Multimodal Large Language Models (MLLMs) achieve strong performance by integrating visual inputs with the rich priors of pretrained language models.<n>However, they often fail on vision-centric tasks, especially when visual evidence conflicts with pretrained knowledge.<n>We explore these failures using two diagnostic paradigms: (1) probing whether visual information is available, via image reconstruction, and (2) measuring multimodal context sensitivity.
  arXiv  Detail & Related papers  (2026-07-28T22:52:34Z)
• Evaluating Vision-Language Models as a Zero-Shot Learning Alternative to You Only Look Once and Optical Character Recognition for Nigerian License Plate Recognition [0.0]
  This study explores the potential of Vision-Language Models (VLMs) as a unified, zeroshot learning solution for Nigerian license plate recognition.<n>We evaluate five selected VLMs: Gemini 2.0 Flash Exp (Google DeepMind), Qwen2.5-VL-7B-Instruct (Alibaba), GPT-4o (OpenAI), Claude 4 Sonnet (Anthropic) and Llama 3.2 Vision 90b (Meta)<n>Results based on Character Error Rate (CER) reveal that Gemini and Qwen significantly outperform other models in both accuracy and robustness, on the challenging image scenarios.
  arXiv  Detail & Related papers  (2026-07-02T10:55:24Z)
• A Multimodal, Multitask System for Generating E Commerce Text Listings from Images [0.0]
  We propose an end to end, multi task system that generates factually grounded textual listings from a single image.<n>The hierarchical generation process proves highly effective, slashing the factual hallucination rate from 12.7% to 7.1%.<n>One minor caveat is that the model does perform 3.5% worse than direct vision-to-language model on ROUGE-L score.
  arXiv  Detail & Related papers  (2025-10-22T11:50:49Z)
• VLMs have Tunnel Vision: Evaluating Nonlocal Visual Reasoning in Leading VLMs [18.349695067647012]
  Visual Language Models excel at complex visual tasks such as VQA and chart understanding, yet recent work suggests they struggle with simple tests.<n>We present an evaluation that tests vision-language models' capacity for nonlocal visual reasoning.<n>Our findings show that despite gains in raw visual acuity, current models lack core visual reasoning capabilities.
  arXiv  Detail & Related papers  (2025-07-04T23:15:52Z)
• HalLoc: Token-level Localization of Hallucinations for Vision Language Models [36.12465376767014]
  Hallucinations pose a significant challenge to the reliability of large vision-language models.<n>HalLoc is a dataset designed for efficient, probabilistic hallucination detection.
  arXiv  Detail & Related papers  (2025-06-12T01:50:35Z)
• VScan: Rethinking Visual Token Reduction for Efficient Large Vision-Language Models [57.2662376527586]
  VScan is a two-stage visual token reduction framework.<n>It addresses token redundancy by: (1) integrating complementary global and local scans with token merging during visual encoding, and (2) introducing pruning at intermediate layers of the language model.<n>VScan achieves a 2.91$times$ speedup in prefilling and a 10$times$ reduction in FLOPs, while retaining 95.4% of the original performance.
  arXiv  Detail & Related papers  (2025-05-28T17:59:08Z)
• MTRE: Multi-Token Reliability Estimation for Hallucination Detection in VLMs [2.6076093883457454]
  Vision-language models (VLMs) now rival human performance on many multimodal tasks, yet they still hallucinate objects or generate unsafe text.<n>We demonstrate that analyzing the complete sequence of early logits potentially provides substantially more diagnostic information.<n>We introduce Multi-Token Reliability Estimation (MTRE), a lightweight, white-box method that aggregates logits from the first ten tokens using multi-token log-likelihood ratios and self-attention.
  arXiv  Detail & Related papers  (2025-05-16T23:00:19Z)
• Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling [78.78822033285938]
  Vision-Language Models (VLMs) excel at visual understanding but often suffer from visual hallucinations.<n>In this work, we introduce REVERSE, a unified framework that integrates hallucination-aware training with on-the-fly self-verification.
  arXiv  Detail & Related papers  (2025-04-17T17:59:22Z)
• Self-Correcting Decoding with Generative Feedback for Mitigating Hallucinations in Large Vision-Language Models [65.4610281589017]
  Large Vision-Language Models (LVLMs) are prone to generating hallucinatory text responses that do not align with the given visual input.<n>We introduce self-correcting Decoding with Generative Feedback (DeGF), a novel training-free algorithm that incorporates feedback from text-to-image generative models into the decoding process.
  arXiv  Detail & Related papers  (2025-02-10T03:43:55Z)
• The Hidden Life of Tokens: Reducing Hallucination of Large Vision-Language Models via Visual Information Steering [42.09744951074433]
  We investigate the internal dynamics of hallucination by examining the tokens logits ranking throughout the generation process.<n>We propose VISTA, a training-free inference-time intervention framework that reduces hallucination while promoting genuine information.
  arXiv  Detail & Related papers  (2025-02-05T21:34:02Z)
• Towards a Systematic Evaluation of Hallucinations in Large-Vision Language Models [57.58426038241812]
  Large Vision-Language Models (LVLMs) have demonstrated remarkable performance in complex multimodal tasks.<n>These models still suffer from hallucinations when required to implicitly recognize or infer diverse visual entities from images.<n>We propose a novel visual question answering (VQA) benchmark that employs contextual reasoning prompts as hallucination attacks.
  arXiv  Detail & Related papers  (2024-12-29T23:56:01Z)
• Aligning Modalities in Vision Large Language Models via Preference Fine-tuning [67.62925151837675]
  In this work, we frame the hallucination problem as an alignment issue, tackle it with preference tuning. Specifically, we propose POVID to generate feedback data with AI models. We use ground-truth instructions as the preferred response and a two-stage approach to generate dispreferred data. In experiments across broad benchmarks, we show that we can not only reduce hallucinations, but improve model performance across standard benchmarks, outperforming prior approaches.
  arXiv  Detail & Related papers  (2024-02-18T00:56:16Z)
• Plausible May Not Be Faithful: Probing Object Hallucination in Vision-Language Pre-training [66.0036211069513]
  Large-scale vision-language pre-trained models are prone to hallucinate non-existent visual objects when generating text. We show that models achieving better scores on standard metrics could hallucinate objects more frequently. Surprisingly, we find that patch-based features perform the best and smaller patch resolution yields a non-trivial reduction in object hallucination.
  arXiv  Detail & Related papers  (2022-10-14T10:27:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.