Identifying and Resolving Pitfalls of Knowledge-Based VQA Benchmarks: Auditing, Repairing, and Augmenting
2026-06-30 • Computation and Language
Computation and LanguageComputer Vision and Pattern RecognitionInformation RetrievalMultimedia
AI summaryⓘ
The authors studied Visual Question Answering systems that use external knowledge to answer questions about images. They found that current tests for these systems often have problems, like unclear questions or missing answers, making accuracy a poor way to measure if the models truly understand and use knowledge. They also noticed that many tests use simple images that don't require complex reasoning. To fix this, the authors created new methods to clean up the tests and add more challenging images, which showed different results when models were re-evaluated. Their work suggests that future tests should focus more on verifying true reasoning instead of just matching answers.
Knowledge-Based Visual Question Answering (KB-VQA)Visual Language Models (VLMs)External Knowledge BaseAnswer DerivabilityQuestion ClarityVisual AmbiguityModel EvaluationBenchmark DatasetReasoning CapabilityVisual-to-Knowledge Mapping
Authors
Qian Ma, S M Rayeed, Charles V. Stewart, Qiong Wu, Yao Ma
Abstract
Knowledge-Based Visual Question Answering (KB-VQA) aims to evaluate whether Visual Language Models (VLMs) can retrieve, ground, and reason over external structured knowledge beyond visual evidence. In practice, answer accuracy is widely adopted as the primary evaluation metric, implicitly treating correctness as a proxy for knowledge-grounded reasoning. However, for existing KB-VQA benchmarks, this proxy relies on critical assumptions that are often overlooked and rendered unreliable by benchmark issues: annotated answer must be derivable from the associated knowledge base, question must be well-posed with sufficient constraints, and visual setting must meaningfully require grounded disambiguation. In this work, we show that these assumptions are systematically violated in existing KB-VQA benchmarks. Our audit reveals substantial instances with missing or contradicted answers and underspecified questions that render accuracy a misleading metric. Furthermore, we find that existing datasets rely on visually trivial, single-entity scenes that bypass the need for sophisticated visual-to-knowledge mapping. We demonstrate that even with controlled architectures, these flaws lead to distorted model rankings and overestimations of reasoning capabilities. To address this, we introduce (1) a principled audit-and-repair protocol that restores answer derivability and question clarity, and (2) a controlled multi-entity augmentation protocol that introduces visual ambiguity to challenge initial retrieval and grounded reasoning. Re-evaluation under corrected and augmented settings yields markedly different performance trends. Our findings call for rethinking evaluation protocols and designing more interaction-aware KB-VQA benchmarks that prioritize verifiable reasoning over simple matching.