Billion-Scale Pretraining with Vision Transformers for Multi-Task Visual Representations

TL;DR

This paper demonstrates that billion-scale pretraining of vision transformers significantly enhances multi-task visual representations, leading to substantial improvements in a real-world visual shopping system.

Contribution

It introduces a billion-image dataset and a systematic weakly-supervised annotation method, replacing CNNs with Transformers for large-scale visual pretraining in industry applications.

Findings

36% improvement in top-1 relevance

23% increase in click-through volume

Transformers outperform CNNs at billion-scale pretraining

Abstract

Large-scale pretraining of visual representations has led to state-of-the-art performance on a range of benchmark computer vision tasks, yet the benefits of these techniques at extreme scale in complex production systems has been relatively unexplored. We consider the case of a popular visual discovery product, where these representations are trained with multi-task learning, from use-case specific visual understanding (e.g. skin tone classification) to general representation learning for all visual content (e.g. embeddings for retrieval). In this work, we describe how we (1) generate a dataset with over a billion images via large weakly-supervised pretraining to improve the performance of these visual representations, and (2) leverage Transformers to replace the traditional convolutional backbone, with insights into both system and performance improvements, especially at 1B+ image scale. To support this backbone model, we detail a systematic approach to deriving weakly-supervised image annotations from heterogenous text signals, demonstrating the benefits of clustering techniques to handle the long-tail distribution of image labels. Through a comprehensive study of offline and online evaluation, we show that large-scale Transformer-based pretraining provides significant benefits to industry computer vision applications. The model is deployed in a production visual shopping system, with 36% improvement in top-1 relevance and 23% improvement in click-through volume. We conduct extensive experiments to better understand the empirical relationships between Transformer-based architectures, dataset scale, and the performance of production vision systems.