Learning Compatible Embeddings

TL;DR

This paper introduces Learning Compatible Embeddings (LCE), a framework that ensures new models are compatible with existing ones in visual retrieval systems by aligning class centers and controlling intra-class distributions, reducing costs and maintaining performance.

Contribution

LCE provides a general, flexible approach for model compatibility in visual retrieval, addressing limitations of previous knowledge distillation methods.

Findings

LCE achieves effective model compatibility across various scenarios.

LCE maintains high accuracy with minimal performance loss.

The framework is applicable to different datasets, loss functions, and architectures.

Abstract

Achieving backward compatibility when rolling out new models can highly reduce costs or even bypass feature re-encoding of existing gallery images for in-production visual retrieval systems. Previous related works usually leverage losses used in knowledge distillation which can cause performance degradations or not guarantee compatibility. To address these issues, we propose a general framework called Learning Compatible Embeddings (LCE) which is applicable for both cross model compatibility and compatible training in direct/forward/backward manners. Our compatibility is achieved by aligning class centers between models directly or via a transformation, and restricting more compact intra-class distributions for the new model. Experiments are conducted in extensive scenarios such as changes of training dataset, loss functions, network architectures as well as feature dimensions, and demonstrate that LCE efficiently enables model compatibility with marginal sacrifices of accuracies. The code will be available at https://github.com/IrvingMeng/LCE.