Saved in:
Bibliographic Details
Main Authors: Moratelli, Nicholas, Davis, Christopher, Ribeiro, Leonardo F. R., Byrne, Bill, Iglesias, Gonzalo
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.12033
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915935957286912
author Moratelli, Nicholas
Davis, Christopher
Ribeiro, Leonardo F. R.
Byrne, Bill
Iglesias, Gonzalo
author_facet Moratelli, Nicholas
Davis, Christopher
Ribeiro, Leonardo F. R.
Byrne, Bill
Iglesias, Gonzalo
contents Large Vision-Language Models (LVLMs) increasingly rely on retrieval to answer knowledge-intensive multimodal questions. Existing benchmarks overlook conflicts between visual and textual evidence and the importance of generating deflections (e.g., Sorry, I cannot answer...) when retrieved knowledge is incomplete. These benchmarks also suffer from rapid obsolescence, as growing LVLM training sets allow models to answer many questions without retrieval. We address these gaps with three contributions. First, we propose a dynamic data curation pipeline that preserves benchmark difficulty over time by filtering for genuinely retrieval-dependent samples. Second, we introduce VLM-DeflectionBench, a benchmark of 2,775 samples spanning diverse multimodal retrieval settings, designed to probe model behaviour under conflicting or insufficient evidence. Third, we define a fine-grained evaluation protocol with four scenarios that disentangle parametric memorization from retrieval robustness. Experiments across 20 state-of-the-art LVLMs indicate that models usually fail to deflect in the presence of noisy or misleading evidence. Our results highlight the need to evaluate not only what models know, but how they behave when they do not, and serve as a reusable and extensible benchmark for reliable KB-VQA evaluation. All resources will be publicly available upon publication.
format Preprint
id arxiv_https___arxiv_org_abs_2604_12033
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Benchmarking Deflection and Hallucination in Large Vision-Language Models
Moratelli, Nicholas
Davis, Christopher
Ribeiro, Leonardo F. R.
Byrne, Bill
Iglesias, Gonzalo
Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
Large Vision-Language Models (LVLMs) increasingly rely on retrieval to answer knowledge-intensive multimodal questions. Existing benchmarks overlook conflicts between visual and textual evidence and the importance of generating deflections (e.g., Sorry, I cannot answer...) when retrieved knowledge is incomplete. These benchmarks also suffer from rapid obsolescence, as growing LVLM training sets allow models to answer many questions without retrieval. We address these gaps with three contributions. First, we propose a dynamic data curation pipeline that preserves benchmark difficulty over time by filtering for genuinely retrieval-dependent samples. Second, we introduce VLM-DeflectionBench, a benchmark of 2,775 samples spanning diverse multimodal retrieval settings, designed to probe model behaviour under conflicting or insufficient evidence. Third, we define a fine-grained evaluation protocol with four scenarios that disentangle parametric memorization from retrieval robustness. Experiments across 20 state-of-the-art LVLMs indicate that models usually fail to deflect in the presence of noisy or misleading evidence. Our results highlight the need to evaluate not only what models know, but how they behave when they do not, and serve as a reusable and extensible benchmark for reliable KB-VQA evaluation. All resources will be publicly available upon publication.
title Benchmarking Deflection and Hallucination in Large Vision-Language Models
topic Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.12033