Vectra: A New Metric, Dataset, and Model for Visual Quality Assessment in E-Commerce In-Image Machine Translation

Qingyu Wu, Yuxuan Han, Haijun Li, Zhao Xu, Jianshan Zhao, Xu Jin, Longyue Wang, Weihua Luo · 2026

Read Vectra: A New Metric, Dataset, and Model for Visual Quality Assessment in E-Commerce In-Image Machine Translation by Qingyu Wu, Yuxuan Han, Haijun Li, Zhao Xu, Jianshan Zhao, Xu Jin, Longyue Wang, Weihua Luo (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-2602-07014.txt.

Description

Qingyu Wu and seven co-authors judge visual quality for e-commerce pictures whose text is translated in place. Translation score alone misses the point: how the rendered picture looks drives whether shoppers engage. Reference-based scores like SSIM lack clear reasons, and model-as-judge scores lack fine, domain-grounded signals. Their system, Vectra, needs no reference picture and is driven by a multimodal model.

Its score splits visual quality into 14 clear dimensions. A defect-area ratio that knows where defects sit cuts annotation doubt. Its data set draws on 1.1 million real product pictures, with a 2,000-picture test set, 30,000 reasoning notes for tuning, and 3,500 expert preference labels. Its 4-billion-parameter model gives both scores and diagnostic reasons, and it lined up with human rankings better than leading commercial models on scoring.

Catalog record

CREATORQingyu Wu, Yuxuan Han, Haijun Li, Zhao Xu, Jianshan Zhao, Xu Jin, Longyue Wang, Weihua Luo
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07014

Browse the full library · About the Tower of Babel