OViP: Online Vision-Language Preference Learning for VLM Hallucination
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866909811722944512 |
|---|---|
| author | Liu, Shujun Wang, Siyuan Li, Zejun Wang, Jianxiang Zeng, Cheng Wei, Zhongyu |
| author_facet | Liu, Shujun Wang, Siyuan Li, Zejun Wang, Jianxiang Zeng, Cheng Wei, Zhongyu |
| contents | Large vision-language models (LVLMs) remain vulnerable to hallucination, often generating content misaligned with visual inputs. Although recent training-based approaches aim to mitigate hallucination, they typically rely on predefined or randomly edited negative samples that do not reflect actual model errors, thus limiting training efficacy. In this work, we propose an Online Vision-language Preference Learning (OViP) framework that dynamically constructs contrastive training data based on the model's own hallucinated outputs. By identifying semantic differences between sampled response pairs and synthesizing negative images using a diffusion model, OViP generates more relevant supervision signals in real time. This failure-driven training enables adaptive alignment of both textual and visual preferences. Moreover, we refine existing evaluation protocols to better capture the trade-off between hallucination suppression and expressiveness. Experiments on hallucination and general benchmarks demonstrate that OViP not only reduces hallucinations while preserving core multi-modal capabilities, but also substantially improves training efficiency. Code is available at https://github.com/lsjlsj35/Online-Vision-Language-Preference-Learning-for-VLM-Hallucination. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_15963 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | OViP: Online Vision-Language Preference Learning for VLM Hallucination Liu, Shujun Wang, Siyuan Li, Zejun Wang, Jianxiang Zeng, Cheng Wei, Zhongyu Computer Vision and Pattern Recognition Computation and Language Large vision-language models (LVLMs) remain vulnerable to hallucination, often generating content misaligned with visual inputs. Although recent training-based approaches aim to mitigate hallucination, they typically rely on predefined or randomly edited negative samples that do not reflect actual model errors, thus limiting training efficacy. In this work, we propose an Online Vision-language Preference Learning (OViP) framework that dynamically constructs contrastive training data based on the model's own hallucinated outputs. By identifying semantic differences between sampled response pairs and synthesizing negative images using a diffusion model, OViP generates more relevant supervision signals in real time. This failure-driven training enables adaptive alignment of both textual and visual preferences. Moreover, we refine existing evaluation protocols to better capture the trade-off between hallucination suppression and expressiveness. Experiments on hallucination and general benchmarks demonstrate that OViP not only reduces hallucinations while preserving core multi-modal capabilities, but also substantially improves training efficiency. Code is available at https://github.com/lsjlsj35/Online-Vision-Language-Preference-Learning-for-VLM-Hallucination. |
| title | OViP: Online Vision-Language Preference Learning for VLM Hallucination |
| topic | Computer Vision and Pattern Recognition Computation and Language |
| url | https://arxiv.org/abs/2505.15963 |