Metric Compatible Training for Online Backfilling in Large-Scale Retrieval

TL;DR

This paper introduces an online backfilling algorithm for large-scale image retrieval that improves performance progressively during model upgrades without additional computational costs.

Contribution

It proposes a metric-compatible contrastive learning approach combined with a distance rank merge technique for efficient online backfilling in retrieval systems.

Findings

Effective performance improvement during backfilling process

No extra computational overhead for merging

Validated on four standard benchmarks

Abstract

Backfilling is the process of re-extracting all gallery embeddings from upgraded models in image retrieval systems. It inevitably requires a prohibitively large amount of computational cost and even entails the downtime of the service. Although backward-compatible learning sidesteps this challenge by tackling query-side representations, this leads to suboptimal solutions in principle because gallery embeddings cannot benefit from model upgrades. We address this dilemma by introducing an online backfilling algorithm, which enables us to achieve a progressive performance improvement during the backfilling process while not sacrificing the final performance of new model after the completion of backfilling. To this end, we first propose a simple distance rank merge technique for online backfilling. Then, we incorporate a reverse transformation module for more effective and efficient merging, which is further enhanced by adopting a metric-compatible contrastive learning approach. These two components help to make the distances of old and new models compatible, resulting in desirable merge results during backfilling with no extra computational overhead. Extensive experiments show the effectiveness of our framework on four standard benchmarks in various settings.