Logit-Attention Divergence: Mitigating Position Bias in Multi-Image Retrieval via Attention-Guided Calibration

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Xian, Mingtao, Yang, Yifeng, Gu, Qinying, Wang, Xinbing, Ye, Nanyang
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911673091096576
author Xian, Mingtao
Yang, Yifeng
Gu, Qinying
Wang, Xinbing
Ye, Nanyang
author_facet Xian, Mingtao
Yang, Yifeng
Gu, Qinying
Wang, Xinbing
Ye, Nanyang
contents Multimodal Large Language Models (MLLMs) have shown strong performance in multi-image cross-modal retrieval, yet suffer from severe position bias, where predictions are dominated by input order rather than semantic relevance. Through empirical analysis, we identify a phenomenon termed Logit-Attention Divergence, in which output logits are heavily biased while internal attention maps remain well-aligned with relevant visual evidence. This observation reveals a fundamental limitation of existing logit-level calibration methods such as PriDe. Based on this insight, we propose a training-free, attention-guided debiasing framework that leverages intrinsic attention signals for instance-level correction at inference time, requiring only a minimal calibration set with negligible computational overhead. Experiments on MS-COCO-based benchmarks show that our method substantially improves permutation invariance and achieves state-of-the-art performance, enhancing accuracy by over 40\% compared to baselines. Code is available at https://github.com/brightXian/LAD.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11591
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Logit-Attention Divergence: Mitigating Position Bias in Multi-Image Retrieval via Attention-Guided Calibration
Xian, Mingtao
Yang, Yifeng
Gu, Qinying
Wang, Xinbing
Ye, Nanyang
Computer Vision and Pattern Recognition
Multimodal Large Language Models (MLLMs) have shown strong performance in multi-image cross-modal retrieval, yet suffer from severe position bias, where predictions are dominated by input order rather than semantic relevance. Through empirical analysis, we identify a phenomenon termed Logit-Attention Divergence, in which output logits are heavily biased while internal attention maps remain well-aligned with relevant visual evidence. This observation reveals a fundamental limitation of existing logit-level calibration methods such as PriDe. Based on this insight, we propose a training-free, attention-guided debiasing framework that leverages intrinsic attention signals for instance-level correction at inference time, requiring only a minimal calibration set with negligible computational overhead. Experiments on MS-COCO-based benchmarks show that our method substantially improves permutation invariance and achieves state-of-the-art performance, enhancing accuracy by over 40\% compared to baselines. Code is available at https://github.com/brightXian/LAD.
title Logit-Attention Divergence: Mitigating Position Bias in Multi-Image Retrieval via Attention-Guided Calibration
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2605.11591