Analyzing and Mitigating Object Hallucination: A Training Bias Perspective

• URL: http://arxiv.org/abs/2508.04567v1
• Date: Wed, 06 Aug 2025 15:51:02 GMT
• Title: Analyzing and Mitigating Object Hallucination: A Training Bias Perspective
• Authors: Yifan Li, Kun Zhou, Wayne Xin Zhao, Lei Fang, Ji-Rong Wen,
• Abstract summary: We introduce a new benchmark, POPEv2, which consists of counterfactual images collected from the training data of LVLMs with certain objects masked.<n>We find that current LVLMs suffer from training bias: they fail to fully leverage their training data and hallucinate more frequently on images seen during training.<n>We propose Obliviate, an efficient and lightweight unlearning method designed to mitigate object hallucination via training bias unlearning.
• Score: 108.09666587800781
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: As scaling up training data has significantly improved the general multimodal capabilities of Large Vision-Language Models (LVLMs), they still suffer from the hallucination issue, generating text that is inconsistent with the visual input. This phenomenon motivates us to systematically investigate the role of training data in hallucination. We introduce a new benchmark, POPEv2, which consists of counterfactual images collected from the training data of LVLMs with certain objects masked. Through comprehensive evaluation on POPEv2, we find that current LVLMs suffer from training bias: they fail to fully leverage their training data and hallucinate more frequently on images seen during training. Specifically, they perform poorly on counterfactual images, often incorrectly answering ``Yes'' to questions about masked objects. To understand this issue, we conduct probing experiments on the models' internal components, revealing that this training bias is primarily located in the language modeling (LM) head. Based on these findings, we propose Obliviate, an efficient and lightweight unlearning method designed to mitigate object hallucination via training bias unlearning. Obliviate identifies the discrepancy between ground-truth labels and model outputs on the training data as a proxy for bias and adopts a parameter- and data-efficient fine-tuning strategy that only updates the LM head. Extensive experiments demonstrate the effectiveness of our approach. While only reusing the training data and updating approximately 2\% of the parameters, Obliviate significantly reduces hallucination across both discriminative and generative tasks. Furthermore, it demonstrates strong scalability with respect to both model size (2B to 72B) and training data volume, and exhibits promising generalization to hallucination types beyond object-level hallucination. Our code and data will be publicly released.

Related papers

• OViP: Online Vision-Language Preference Learning [26.54737360667123]
  Large vision-language models (LVLMs) remain vulnerable to hallucination, often generating content misaligned with visual inputs.<n>We propose an Online Vision-language Preference Learning framework that dynamically constructs contrastive training data based on the model's own hallucinated outputs.<n>Experiments on hallucination and general benchmarks demonstrate that OViP effectively reduces hallucinations while preserving core multi-modal capabilities.
  arXiv  Detail & Related papers  (2025-05-21T19:26:09Z)
• HalluLens: LLM Hallucination Benchmark [49.170128733508335]
  Large language models (LLMs) often generate responses that deviate from user input or training data, a phenomenon known as "hallucination"<n>This paper introduces a comprehensive hallucination benchmark, incorporating both new extrinsic and existing intrinsic evaluation tasks.
  arXiv  Detail & Related papers  (2025-04-24T13:40:27Z)
• FiVL: A Framework for Improved Vision-Language Alignment through the Lens of Training, Evaluation and Explainability [10.184567639685321]
  We introduce FiVL, a novel method for constructing datasets designed to train LVLMs for enhanced visual grounding.<n>We present benchmarks to assess the model's ability to use image as substantive evidence.<n>We identify attention heads with the strongest vision-language alignment, enabling explainability on visual-driven hallucinations.
  arXiv  Detail & Related papers  (2024-12-19T09:24:10Z)
• Pre-Training Multimodal Hallucination Detectors with Corrupted Grounding Data [4.636499986218049]
  Multimodal language models can exhibit hallucinations in their outputs, which limits their reliability. We propose an approach to improve the sample efficiency of these models by creating corrupted grounding data.
  arXiv  Detail & Related papers  (2024-08-30T20:11:00Z)
• Training Language Models on the Knowledge Graph: Insights on Hallucinations and Their Detectability [83.0884072598828]
  Hallucinations come in many forms, and there is no universally accepted definition. We focus on studying only those hallucinations where a correct answer appears verbatim in the training set. We find that for a fixed dataset, larger and longer-trained LMs hallucinate less. While we see detector size improves performance on fixed LM's outputs, we find an inverse relationship between the scale of the LM and the detectability of its hallucinations.
  arXiv  Detail & Related papers  (2024-08-14T23:34:28Z)
• The First to Know: How Token Distributions Reveal Hidden Knowledge in Large Vision-Language Models? [34.27319941609499]
  This study uses linear probing to shed light on the hidden knowledge at the output layers of LVLMs. We demonstrate that the logit distributions of the first tokens contain sufficient information to determine whether to respond to the instructions.
  arXiv  Detail & Related papers  (2024-03-14T02:25:35Z)
• Less is More: Mitigating Multimodal Hallucination from an EOS Decision Perspective [55.41815486466186]
  Large Multimodal Models (LMMs) often suffer from multimodal hallucinations, wherein they create content that is not present in the visual inputs. In this paper, we explore a new angle of this issue: overly detailed training data hinders the model's ability to timely terminate generation. We find that the model assesses the completeness of the entire sequence by comparing the generated text with the image.
  arXiv  Detail & Related papers  (2024-02-22T13:33:13Z)
• 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)
• HalluciDoctor: Mitigating Hallucinatory Toxicity in Visual Instruction Data [102.56792377624927]
  hallucinations inherent in machine-generated data remain under-explored. We present a novel hallucination detection and elimination framework, HalluciDoctor, based on the cross-checking paradigm. Our method successfully mitigates 44.6% hallucinations relatively and maintains competitive performance compared to LLaVA.
  arXiv  Detail & Related papers  (2023-11-22T04:52:58Z)

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.