V$^2$L: Leveraging Vision and Vision-language Models into Large-scale Product Retrieval

TL;DR

This paper presents a top-performing ensemble approach for large-scale product retrieval in ecommerce, combining vision and vision-language models with a coarse-to-fine training strategy to improve accuracy.

Contribution

It introduces a novel ensemble method that leverages the complementarity of vision and vision-language models with a two-stage training pipeline for product retrieval.

Findings

Achieved 0.7623 MAR@10, first place in eBay eProduct Visual Search Challenge.

Demonstrated benefits of combining vision and vision-language models.

Implemented a coarse-to-fine metric learning approach for improved retrieval.

Abstract

Product retrieval is of great importance in the ecommerce domain. This paper introduces our 1st-place solution in eBay eProduct Visual Search Challenge (FGVC9), which is featured for an ensemble of about 20 models from vision models and vision-language models. While model ensemble is common, we show that combining the vision models and vision-language models brings particular benefits from their complementarity and is a key factor to our superiority. Specifically, for the vision models, we use a two-stage training pipeline which first learns from the coarse labels provided in the training set and then conducts fine-grained self-supervised training, yielding a coarse-to-fine metric learning manner. For the vision-language models, we use the textual description of the training image as the supervision signals for fine-tuning the image-encoder (feature extractor). With these designs, our solution achieves 0.7623 MAR@10, ranking the first place among all the competitors. The code is available at: \href{https://github.com/WangWenhao0716/V2L}{V$^2$L}.