发现主流视觉语言基准存在固有偏差，可能误导模型评估
A Good CREPE needs more than just Sugar: Investigating Biases in Compositional Vision-Language Benchmarks
- 分析17个常用基准的设计缺陷，揭示数据构造中的系统性偏差
- 简单启发式方法表现媲美CLIP模型，说明基准无法真实衡量组合理解能力
- 提出构建更鲁棒基准的建议，适合评估与设计视觉语言模型的研究者
我们研究了17个常用于评估视觉语言模型（VLMs）组合理解能力的基准（如SugarCREPE、VALSE），审视其数据来源（如MS-COCO）和筛选流程（如构造负样本图像/描述），发现大多数基准存在内在偏差。研究发现，仅依赖词长或语言模型对数似然等盲启发式方法的表现可与CLIP模型相当，表明这些基准未能有效测量真正的组合理解能力。我们证明根本原因在于正负样本间分布不对称，由基准构建过程导致。为此，我们提出若干关键建议，以构建更稳健、不易被简单策略攻破的视觉语言组合理解基准。
原文摘要 · Abstract (English)
We investigate 17 benchmarks (e.g. SugarCREPE, VALSE) commonly used for measuring compositional understanding capabilities of vision-language models (VLMs). We scrutinize design choices in their construction, including data source (e.g. MS-COCO) and curation procedures (e.g. constructing negative images/captions), uncovering several inherent biases across most benchmarks. We find that blind heuristics (e.g. token-length, log-likelihood under a language model) perform on par with CLIP models, indicating that these benchmarks do not effectively measure compositional understanding. We demonstrate that the underlying factor is a distribution asymmetry between positive and negative images/captions, induced by the benchmark construction procedures. To mitigate these issues, we provide a few key recommendations for constructing more robust vision-language compositional understanding benchmarks, that would be less prone to such simple attacks.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。