English

Probing Visual Language Priors in VLMs

Computer Vision and Pattern Recognition 2025-04-15 v4 Machine Learning

Abstract

Despite recent advances in Vision-Language Models (VLMs), they may over-rely on visual language priors existing in their training data rather than true visual reasoning. To investigate this, we introduce ViLP, a benchmark featuring deliberately out-of-distribution images synthesized via image generation models and out-of-distribution Q&A pairs. Each question in ViLP is coupled with three potential answers and three corresponding images: one that can be resolved by text priors alone and two that demand visual reasoning. Although, humans achieve near-perfect accuracy, modern VLMs falter; for instance, GPT-4 achieves only 66.17% on ViLP. To alleviate this, we propose a self-improving framework in which models generate new VQA data, then apply pixel-level and semantic corruptions to form "good-bad" image pairs for self-training. Our training objectives compel VLMs to focus more on the actual visual inputs, and we demonstrate their effectiveness in boosting the performance of open-source VLMs, including LLaVA-v1.5 and Cambrian.

Keywords

Cite

@article{arxiv.2501.00569,
  title  = {Probing Visual Language Priors in VLMs},
  author = {Tiange Luo and Ang Cao and Gunhee Lee and Justin Johnson and Honglak Lee},
  journal= {arXiv preprint arXiv:2501.00569},
  year   = {2025}
}

Comments

Project Page: https://vilp-team.github.io/

R2 v1 2026-06-28T20:53:32.979Z