Cross-Image Contrastive Decoding: Precise, Lossless Suppression of Language Priors in Large Vision-Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhao, Jianfei, Zhang, Feng, Sun, Xin, Kong, Lingxing, Tan, Zhixing, Feng, Chong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909790961139712
author Zhao, Jianfei
Zhang, Feng
Sun, Xin
Kong, Lingxing
Tan, Zhixing
Feng, Chong
author_facet Zhao, Jianfei
Zhang, Feng
Sun, Xin
Kong, Lingxing
Tan, Zhixing
Feng, Chong
contents Over-reliance on language priors is a major cause of hallucinations in Large Vision-Language Models (LVLMs), often leading to outputs that are linguistically plausible but visually inconsistent. Recent studies have explored contrastive decoding as a training-free solution. However, these methods typically construct contrastive visual inputs by perturbing the original image, resulting in distorted contrastive distributions, incomplete contrastive signals, and excessive suppression of language priors. Motivated by the observation that language priors tend to remain consistent across different images, we propose Cross-Image Contrastive Decoding (CICD), a simple yet effective training-free method that uses unrelated images as contrastive visual inputs. To address the issue of over-suppressing language priors, which can negatively affect the quality of generated responses, we further introduce a dynamic selection mechanism based on the cross-image differences in model behavior. By selectively suppressing language priors, our method reduces hallucinations without compromising the model's performance. Extensive experiments across multiple benchmarks and LVLMs confirm the effectiveness and generalizability of CICD, particularly in image captioning, where language priors are especially dominant.
format Preprint
id arxiv_https___arxiv_org_abs_2505_10634
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cross-Image Contrastive Decoding: Precise, Lossless Suppression of Language Priors in Large Vision-Language Models
Zhao, Jianfei
Zhang, Feng
Sun, Xin
Kong, Lingxing
Tan, Zhixing
Feng, Chong
Computer Vision and Pattern Recognition
Over-reliance on language priors is a major cause of hallucinations in Large Vision-Language Models (LVLMs), often leading to outputs that are linguistically plausible but visually inconsistent. Recent studies have explored contrastive decoding as a training-free solution. However, these methods typically construct contrastive visual inputs by perturbing the original image, resulting in distorted contrastive distributions, incomplete contrastive signals, and excessive suppression of language priors. Motivated by the observation that language priors tend to remain consistent across different images, we propose Cross-Image Contrastive Decoding (CICD), a simple yet effective training-free method that uses unrelated images as contrastive visual inputs. To address the issue of over-suppressing language priors, which can negatively affect the quality of generated responses, we further introduce a dynamic selection mechanism based on the cross-image differences in model behavior. By selectively suppressing language priors, our method reduces hallucinations without compromising the model's performance. Extensive experiments across multiple benchmarks and LVLMs confirm the effectiveness and generalizability of CICD, particularly in image captioning, where language priors are especially dominant.
title Cross-Image Contrastive Decoding: Precise, Lossless Suppression of Language Priors in Large Vision-Language Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2505.10634