From Pixels to Facts (Pix2Fact): Benchmarking Multi-Hop Reasoning for Fine-Grained Visual Fact Checking

Yifan Jiang, Cong Zhang, Bofei Zhang, Yifan Yang, Bingzhang Wang, Yew-Soon Ong · 2026

Read From Pixels to Facts (Pix2Fact): Benchmarking Multi-Hop Reasoning for Fine-Grained Visual Fact Checking by Yifan Jiang, Cong Zhang, Bofei Zhang, Yifan Yang, Bingzhang Wang, Yew-Soon Ong (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00593.txt.

Description

Yifan Jiang and colleagues introduce Pix2Fact, a visual question-answering benchmark testing expert perception plus knowledge search, areas existing benchmarks judge alone. Its 1,000 4K images span eight scenarios, with questions crafted by PhD-holding annotators across disciplines, each demanding visual grounding and external knowledge. Testing ten state-of-the-art vision-language models, the best reaches only 51.7% accuracy even with visual ground truth and search tools. Failures trace to grounding errors, shallow search use, and trouble retrieving long-tail, unstructured local information, exposing limits in real visual comprehension.

Catalog record

CREATORYifan Jiang, Cong Zhang, Bofei Zhang, Yifan Yang, Bingzhang Wang, Yew-Soon Ong
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61394
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00593

Browse the full library · About the Tower of Babel