🤖 AI Summary
This work addresses the challenge of scalable product matching in large-scale retail settings, where significant domain discrepancies exist between high-resolution catalog images and real in-store photographs. To bridge this domain gap, the authors propose Cat2Real, a multi-stage contrastive learning framework that reformulates product identification as a cross-domain embedding retrieval task. The approach integrates hierarchical hard negative mining guided by both item-level and image-level similarity, cross-domain contrastive losses, and visual embedding learning, enabling zero-shot generalization to novel products without requiring real-world training data for new items. Experimental results demonstrate that the model substantially improves matching accuracy and system scalability on unseen products and categories, effectively narrowing the domain gap between catalog and real-world imagery.
📝 Abstract
Automated product recognition is a cornerstone of modern retail intelligence; however, accurately matching real-world, in-store images against extensive corporate catalogs remains a major scalability bottleneck for large-scale applications. In this work, we address this challenge by reformulating the task as an embedding-based cross-domain retrieval problem rather than a standard closed-set classification task. Specifically, we define the objective as retrieving the most corresponding catalog reference image for a given real-world product query crop from an expansive inventory. To bridge the severe domain gap between pristine studio packshots and noisy in-store queries, we introduce a novel catalog-to-real multi-stage contrastive learning paradigm (Cat2Real). This framework fine-tunes a vision backbone by systematically exploiting both item-level and image-level similarities to drive targeted hard negative mining. Extensive empirical evaluations demonstrate that our paradigm scales seamlessly to unseen products and categories, yielding outstanding zero-shot generalization performance even in the complete absence of real-world training images for novel inventory.