E-commerce product understanding demands by nature, strong multimodal comprehension from text, images, and structured attributes. General-purpose Vision-Language Models (VLMs) enable generalizable multimodal latent modelling, yet there is no documented, well-known strategy for adapting them to the attribute-centric, multi-image, and noisy nature of e-commerce data, without sacrificing general performance. In this work, we show through a large-scale experimental study, how targeted adaptation of general VLMs can substantially improve e-commerce performance while preserving broad multimodal capabilities. Furthermore, we propose a novel extensive evaluation suite covering deep product understanding, strict instruction following, and dynamic attribute extraction.
@article{arxiv.2602.11733,
title = {Adapting Vision-Language Models for E-commerce Understanding at Scale},
author = {Matteo Nulli and Vladimir Orshulevich and Tala Bazazo and Christian Herold and Michael Kozielski and Marcin Mazur and Szymon Tuzel and Cees G. M. Snoek and Seyyed Hadi Hashemi and Omar Javed and Yannick Versley and Shahram Khadivi},
journal= {arXiv preprint arXiv:2602.11733},
year = {2026}
}