cs.CVNov 17, 2026

M3^3-Test: Coverage Gaps in Joint-Perturbation Testing of Visual Question Answering

Authors: Enhui Wan, Xiaowan Wang

Organizations: Beijing University of Posts and Telecommunications, Shahe Campus, No. 1 Nanfeng Road, Shahe Higher Education Park, Changping District, Beijing, 102206, China

Abstract

A visual question answering (VQA) model can pass every image-only and question-only test yet fail when the same changes occur together. Standard robustness reports average over conditions and therefore cannot say which accepted requests remain vulnerable. We introduce M3-Test, a coverage audit that aligns each clean input with its single-modality variants and a complete 5x5 family, 4x4 strength joint library, so that any reduced test suite can be replayed on the same requests. Across 2.6 million responses from four vision-language models, joint failures affect 26-29% of the Qwen questions that pass all 40 single-modality tests, although fewer than 0.6% of their joint records fail. Passing the 25 maximum-strength corners does not close this gap: 11-14% of the accepted questions still fail elsewhere in the library. Re-executing the archived failures reproduces about 90% of them, against a 0.3-0.6% flip rate among matched passing controls, and re-checking the most fragile failure records supports 80-89% of them. On the evidence examined, neither execution noise nor grading error can fully account for the gap. The same aligned grids make previous evaluations reusable. On held-out images, a risk order learned from other models discovers about 80% of affected Qwen questions within 25 additional queries, compared with 65-68% for maximum-first and 53-57% for pairwise covering arrays. M3-Test thus turns a passing robustness suite into a request-level account of what remains untested, supporting acceptance, regression review and test allocation.

Figures & tables

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

May 19, 2026

MetaRA: Metamorphic Robustness Assessment for Multimodal Large Language Model-based Visual Question Answering Systems

Visual Question Answering (VQA), as the representative multimodal task, serves as a key benchmark for evaluating the reasoning capabilities of Multimodal Large Language Models (MLLMs). However, existing evaluations largely rely on static datasets and accuracy-based metrics, which fail to capture robustness, consistency, and generalization. Inspired by Metamorphic Testing (MT), we propose Metamorphic Robustness Assessment (MetaRA), a testing framework that employs Metamorphic Relations (MRs) to systematically probe vulnerabilities in MLLM-based VQA systems. MetaRA generates controlled variations of image-question inputs based on specific MRs and evaluates models across diverse conditions. Applying MetaRA to multiple MLLM-based VQA models across different tasks reveals nuanced failure patterns, including sensitivity to linguistic perturbations, over-reliance on superficial visual cues, and deeper weaknesses in multimodal reasoning. Experimental results demonstrate that MetaRA provides richer diagnostic insights than conventional accuracy metrics, exposing failure modes that remain hidden under standard benchmarks. Overall, this work highlights the need for systematic robustness evaluation in VQA and positions metamorphic assessment as a scalable, model-agnostic approach toward trustworthy multimodal AI.
Oct 22, 2026

VLTest: Finding Joint Image-Text Failures in Vision-Language Models

Vision-language models (VLMs) are increasingly embedded in software systems. Testing them raises two problems. Mutating either input can change the expected answer, so an output change is not by itself a failure. Some failures appear only when both inputs are mutated; attributing them to the combination requires testing each mutation alone. We present VLTest, a black-box VLM testing approach. It generates image variants by editing VQGAN latent codes and text variants with a rewrite-expand-prune pipeline, filters variants likely to change the expected answer, and tests each variant alone before testing combinations, so every joint failure has four matched cells. A VLM judge confirms a failure only after checking that the input keeps the expected answer. On three tasks with eight VLMs, including two commercial API models, the oracle confirms 81.7% of the candidates, with 88.8% to 94.6% precision against human labels, and four-cell confirmation rejects 22.5% of joint candidates. Under the same budget, VLTest's generators find 1.8 to 2.5 times as many confirmed failures as SINVAD and TextFooler, and its joint tests find 91 to 197 attributed joint failures per 1,000 joint calls, 1.3 times as many as the same pairing applied to the baselines' variants.
May 20, 2026

VISTAQA: Benchmarking Joint Visual Question Answering and Pixel-Level Evidence

Establishing a clear link between model predictions and the visual evidence that supports them is critical for transparency and reliability in multimodal reasoning, yet current multimodal large language model (MLLM) evaluations do not explicitly enforce this alignment. Existing benchmarks assess either textual answer correctness or pixel-level localization in isolation, leaving the coupling of reasoning and grounding an open challenge. We introduce VISTAQA, a comprehensive benchmark for joint evaluation of free-form answer correctness and pixel-level evidence grounding in visual question answering. VISTAQA comprises 1,157 expert-curated samples spanning six task types and six visual domains, ranging from direct perception to compositional and relational reasoning. VISTAQA requires models to not only answer correctly, but to also provide precise segmentation masks that support their answers. It also includes hallucination-aware examples where no valid visual evidence exists. To support this enhanced evaluation, we introduce GROVE, a unified evaluation metric that enforces joint correctness by combining textual accuracy and grounding quality via a per-sample geometric mean, ensuring neither dimension can compensate for deficiencies in the other. Comprehensive experiments across grounding-aware models and hybrid pipelines with general-purpose MLLMs reveal that even the strongest systems achieve limited performance under GROVE, highlighting a substantial gap between answer accuracy and visual evidence alignment. Project page can be found at https://vistaqa.github.io.