Efficient Large-Scale Visual Representation Learning And Evaluation
This work addresses efficiency challenges in visual representation learning for large-scale e-commerce platforms, but it appears incremental as it builds on existing architectures and methods.
The paper tackles the problem of efficiently learning visual representations for large-scale e-commerce recommendations by comparing pretrained CNN and ViT backbones, and it introduces a novel multilingual text-to-image generative offline evaluation method, with results from deployed systems.
Efficiently learning visual representations of items is vital for large-scale recommendations. In this article we compare several pretrained efficient backbone architectures, both in the convolutional neural network (CNN) and in the vision transformer (ViT) family. We describe challenges in e-commerce vision applications at scale and highlight methods to efficiently train, evaluate, and serve visual representations. We present ablation studies evaluating visual representations in several downstream tasks. To this end, we present a novel multilingual text-to-image generative offline evaluation method for visually similar recommendation systems. Finally, we include online results from deployed machine learning systems in production on a large scale e-commerce platform.